Thread: Add docs stub for recovery.conf
Hi all
I noticed that when recovery.conf was removed in 2dedf4d9a8 (yay!) the docs for it were removed completely as well. That's largely sensible, but is confusing when users have upgraded and are trying to find out what happened, or how to configure equivalent functionality.
https://www.postgresql.org/docs/11/recovery-config.html just vanished for /12/, and as a result https://www.postgresql.org/docs/current/recovery-config.html is a 404. I think that's unhelpful since we encourage people to use /current/ links.
The attached patch restores the recovery-config.html page with a short note explaining why it's gone and what to do instead. It's added to a new appendix "Obsolete or renamed features, settings and files".
I found it remarkably hard to find out what exactly made a "standby server" actually be a standby server in the docs so I have added a couple of cross-references that make the role of the standby.signal file much more discoverable from relevant locations.
I propose a policy that we preserve our <chapter> and <sect1> ids. We should move them to an "obsolete" section at the end, like the one I created here, and provide stubs for them instead of removing them. That'll help prevent us from breaking links on the wider web, in 3rd party documentation, etc.
Attachment
On Tue, Nov 10, 2020 at 01:38:14PM +0800, Craig Ringer wrote: > Hi all > > I noticed that when recovery.conf was removed in 2dedf4d9a8 (yay!) the docs for > it were removed completely as well. That's largely sensible, but is confusing > when users have upgraded and are trying to find out what happened, or how to > configure equivalent functionality. I don't see the logic in carrying doc stuff that we don't have anymore. -- Bruce Momjian <bruce@momjian.us> https://momjian.us EnterpriseDB https://enterprisedb.com The usefulness of a cup is in its emptiness, Bruce Lee
On Wed, Nov 11, 2020 at 12:44 PM Bruce Momjian <bruce@momjian.us> wrote:
On Tue, Nov 10, 2020 at 01:38:14PM +0800, Craig Ringer wrote:
> Hi all
>
> I noticed that when recovery.conf was removed in 2dedf4d9a8 (yay!) the docs for
> it were removed completely as well. That's largely sensible, but is confusing
> when users have upgraded and are trying to find out what happened, or how to
> configure equivalent functionality.
I don't see the logic in carrying doc stuff that we don't have anymore.
I do. Saying why something went away has value. For small stuff you have commit messages. For user-facing documentation stuff that warranted its own page, having said page remain and describe the change seems worthwhile.
David J.
> On 11 Nov 2020, at 20:44, Bruce Momjian <bruce@momjian.us> wrote:
> On Tue, Nov 10, 2020 at 01:38:14PM +0800, Craig Ringer wrote:
>> I noticed that when recovery.conf was removed in 2dedf4d9a8 (yay!) the docs for
>> it were removed completely as well. That's largely sensible, but is confusing
>> when users have upgraded and are trying to find out what happened, or how to
>> configure equivalent functionality.
>
> I don't see the logic in carrying doc stuff that we don't have anymore.
Well, we do have that already in <tip>'s sprinkled across the docs where it
makes sense to help transitioning users, like this one in func.sgml:
"Prior to PostgreSQL 12, it was possible to skip arbitrary text in the
input string using non-letter or non-digit characters..."
It doesn't seem like a terrible idea to do a similar one for recovery.conf.
cheers ./daniel
On Wed, Nov 11, 2020 at 08:59:40PM +0100, Daniel Gustafsson wrote: > > On 11 Nov 2020, at 20:44, Bruce Momjian <bruce@momjian.us> wrote: > > On Tue, Nov 10, 2020 at 01:38:14PM +0800, Craig Ringer wrote: > > >> I noticed that when recovery.conf was removed in 2dedf4d9a8 (yay!) the docs for > >> it were removed completely as well. That's largely sensible, but is confusing > >> when users have upgraded and are trying to find out what happened, or how to > >> configure equivalent functionality. > > > > I don't see the logic in carrying doc stuff that we don't have anymore. > > Well, we do have that already in <tip>'s sprinkled across the docs where it > makes sense to help transitioning users, like this one in func.sgml: > > "Prior to PostgreSQL 12, it was possible to skip arbitrary text in the > input string using non-letter or non-digit characters..." > > It doesn't seem like a terrible idea to do a similar one for recovery.conf. I am fine with a tip. The patch looked like it was creating a new chapter for it. -- Bruce Momjian <bruce@momjian.us> https://momjian.us EnterpriseDB https://enterprisedb.com The usefulness of a cup is in its emptiness, Bruce Lee
> On 11 Nov 2020, at 21:01, Bruce Momjian <bruce@momjian.us> wrote: > > On Wed, Nov 11, 2020 at 08:59:40PM +0100, Daniel Gustafsson wrote: >>> On 11 Nov 2020, at 20:44, Bruce Momjian <bruce@momjian.us> wrote: >>> On Tue, Nov 10, 2020 at 01:38:14PM +0800, Craig Ringer wrote: >> >>>> I noticed that when recovery.conf was removed in 2dedf4d9a8 (yay!) the docs for >>>> it were removed completely as well. That's largely sensible, but is confusing >>>> when users have upgraded and are trying to find out what happened, or how to >>>> configure equivalent functionality. >>> >>> I don't see the logic in carrying doc stuff that we don't have anymore. >> >> Well, we do have that already in <tip>'s sprinkled across the docs where it >> makes sense to help transitioning users, like this one in func.sgml: >> >> "Prior to PostgreSQL 12, it was possible to skip arbitrary text in the >> input string using non-letter or non-digit characters..." >> >> It doesn't seem like a terrible idea to do a similar one for recovery.conf. > > I am fine with a tip. The patch looked like it was creating a new > chapter for it. I admittedly hadn't looked at the patch, but now that I have I agree with not adding a separate "obsolete" topic for it. I'd prefer to use tips within the docs, they will also help guide users who search for recovery.conf and lands in the tip which is next to the relevant updated documentation on the topic. cheers ./daniel
On Thu, Nov 12, 2020 at 3:44 AM Bruce Momjian <bruce@momjian.us> wrote:
On Tue, Nov 10, 2020 at 01:38:14PM +0800, Craig Ringer wrote:
> Hi all
>
> I noticed that when recovery.conf was removed in 2dedf4d9a8 (yay!) the docs for
> it were removed completely as well. That's largely sensible, but is confusing
> when users have upgraded and are trying to find out what happened, or how to
> configure equivalent functionality.
I don't see the logic in carrying doc stuff that we don't have anymore.
I explained why.
Here's how the rendered docs look: https://imgur.com/a/VyjzEw5
Think. You're used to recovery.conf. You've recently switched to pg 12. You search for "recovery.conf" or "primary_slot_name" or "standby_mode" or something. You of course land up at https://www.postgresql.org/docs/11/recovery-config.html or another version.
What do you do now? There's no "12" or "current" link for your version. You don't follow postgres development closely, and you have no idea we removed the file. You might work it out from the release notes. But even then, if you try searching for "standby_mode" in the updated docs will tell you basically nothing, it has just vanished
Also by simply deleting the page, we've broken web links to https://www.postgresql.org/docs/current/recovery-config.html
So there are three really good reasons:
* Help users of the web docs navigate to the right place when we remove things
* Don't break web links (breaking links without redirects is bad, the web is sad)
* Help upgrading users who know the old terms find the new terms
I'd welcome your suggestions on other ways to arrange this, so long as it meets the basic requirement "retain the linktable target 'recovery-config' "
In general I think it's quite unpleasant for users to have docs sections just vanish. I strongly suggest that we enact a policy going forwards that any <chapter> or <sect1> removal should be accompanied by a redirect or stub that helps users find the new contents. It regularly annoys me when I'm trying to navigate around various versions of the docs and things just vanish.
On Thu, Nov 12, 2020 at 4:01 AM Bruce Momjian <bruce@momjian.us> wrote:
On Wed, Nov 11, 2020 at 08:59:40PM +0100, Daniel Gustafsson wrote:
> > On 11 Nov 2020, at 20:44, Bruce Momjian <bruce@momjian.us> wrote:
> > On Tue, Nov 10, 2020 at 01:38:14PM +0800, Craig Ringer wrote:
>
> >> I noticed that when recovery.conf was removed in 2dedf4d9a8 (yay!) the docs for
> >> it were removed completely as well. That's largely sensible, but is confusing
> >> when users have upgraded and are trying to find out what happened, or how to
> >> configure equivalent functionality.
> >
> > I don't see the logic in carrying doc stuff that we don't have anymore.
>
> Well, we do have that already in <tip>'s sprinkled across the docs where it
> makes sense to help transitioning users, like this one in func.sgml:
>
> "Prior to PostgreSQL 12, it was possible to skip arbitrary text in the
> input string using non-letter or non-digit characters..."
>
> It doesn't seem like a terrible idea to do a similar one for recovery.conf.
I am fine with a tip. The patch looked like it was creating a new
chapter for it.
It is. Or rather, an appendix right at the end to hold info on things we removed or renamed and where to find them now.
You can't AFAICS make docbook create a toplevel linkable file for a <tip> . A <tip> won't un-break https://www.postgresql.org/docs/current/recovery-config.html or make help people who visit https://www.postgresql.org/docs/11/recovery-config.html figure out what's going on if they're using pg12 and there's no link to version 12 in the nav section. A <tip> won't add index entries for renamed settings, so someone looking up "standby_mode" can find out that we've switched to a file called "standby.signal" instead.
Pretend you're a user who has upgraded from pg 11. You're looking at the Pg 12 docs. How long does it take you to find out how to make a server into a standby now? It took me longer than I would've expected...
On Thu, Nov 12, 2020 at 10:21:02AM +0800, Craig Ringer wrote: > Here's how the rendered docs look: https://imgur.com/a/VyjzEw5 > > Think. You're used to recovery.conf. You've recently switched to pg 12. You > search for "recovery.conf" or "primary_slot_name" or "standby_mode" or > something. You of course land up at https://www.postgresql.org/docs/11/ > recovery-config.html or another version. > > What do you do now? There's no "12" or "current" link for your version. You > don't follow postgres development closely, and you have no idea we removed the > file. You might work it out from the release notes. But even then, if you try > searching for "standby_mode" in the updated docs will tell you basically > nothing, it has just vanished > > Also by simply deleting the page, we've broken web links to https:// > www.postgresql.org/docs/current/recovery-config.html > > So there are three really good reasons: > > * Help users of the web docs navigate to the right place when we remove things > * Don't break web links (breaking links without redirects is bad, the web is > sad) > * Help upgrading users who know the old terms find the new terms > > I'd welcome your suggestions on other ways to arrange this, so long as it meets > the basic requirement "retain the linktable target 'recovery-config' " This is certainly not the first or last time we will rename things. Fortunately, this has already been discussed in the renaming of default roles to predefined roles: https://www.postgresql.org/message-id/flat/157742545062.1149.11052653770497832538%40wrigleys.postgresql.org This naming change has not happened yet, even though the issue is 11 months old, but the agreed-upon way to handle this is to use a website redirect that links to the new text. You can add a "tip" there so they understand the renaming has happened. -- Bruce Momjian <bruce@momjian.us> https://momjian.us EnterpriseDB https://enterprisedb.com The usefulness of a cup is in its emptiness, Bruce Lee
Greetings, * Bruce Momjian (bruce@momjian.us) wrote: > On Thu, Nov 12, 2020 at 10:21:02AM +0800, Craig Ringer wrote: > > Here's how the rendered docs look: https://imgur.com/a/VyjzEw5 > > > > Think. You're used to recovery.conf. You've recently switched to pg 12. You > > search for "recovery.conf" or "primary_slot_name" or "standby_mode" or > > something. You of course land up at https://www.postgresql.org/docs/11/ > > recovery-config.html or another version. > > > > What do you do now? There's no "12" or "current" link for your version. You > > don't follow postgres development closely, and you have no idea we removed the > > file. You might work it out from the release notes. But even then, if you try > > searching for "standby_mode" in the updated docs will tell you basically > > nothing, it has just vanished > > > > Also by simply deleting the page, we've broken web links to https:// > > www.postgresql.org/docs/current/recovery-config.html > > > > So there are three really good reasons: > > > > * Help users of the web docs navigate to the right place when we remove things > > * Don't break web links (breaking links without redirects is bad, the web is > > sad) > > * Help upgrading users who know the old terms find the new terms > > > > I'd welcome your suggestions on other ways to arrange this, so long as it meets > > the basic requirement "retain the linktable target 'recovery-config' " > > This is certainly not the first or last time we will rename things. Indeed, we've renamed things a number of times before, eg: https://www.postgresql.org/docs/current/pgwaldump.html where the 9.6 link goes to: https://www.postgresql.org/docs/9.6/pgxlogdump.html and the 'current' link from the 9.6 page goes to the pgwaldump page, which all works pretty well, if all you're looking at is our documentation and not considering external links into the documentation. However, that isn't what Craig's raising a concern over here (at least, not exclusively), it's this issue: https://www.postgresql.org/docs/current/pgxlogdump.html Which currently goes to a 404. Now, the pgweb feature that Jonathan wrote recently might actually be exactly what we need to fix that, and to address the issue with recovery config documentation that Craig raises. > Fortunately, this has already been discussed in the renaming of default > roles to predefined roles: > > https://www.postgresql.org/message-id/flat/157742545062.1149.11052653770497832538%40wrigleys.postgresql.org > > This naming change has not happened yet, even though the issue is 11 > months old, but the agreed-upon way to handle this is to use a website > redirect that links to the new text. You can add a "tip" there so they > understand the renaming has happened. That rename will suffer the same problem that Craig is concerned about here regarding the 'current' link, once it's done. I tend to agree with Craig that it'd be good to improve on this situation, and I've reached out to Jonathan to ask about using his new feature to have those /current/ links redirect to the renamed page. I'm actually wondering if maybe we should just always do that for all the dog page aliases.. Might make more sense to discuss this over on -www though. Thanks, Stephen
Attachment
Greetings, * Stephen Frost (sfrost@snowman.net) wrote: > * Bruce Momjian (bruce@momjian.us) wrote: > > On Thu, Nov 12, 2020 at 10:21:02AM +0800, Craig Ringer wrote: > > > Here's how the rendered docs look: https://imgur.com/a/VyjzEw5 > > > > > > Think. You're used to recovery.conf. You've recently switched to pg 12. You > > > search for "recovery.conf" or "primary_slot_name" or "standby_mode" or > > > something. You of course land up at https://www.postgresql.org/docs/11/ > > > recovery-config.html or another version. > > > > > > What do you do now? There's no "12" or "current" link for your version. You > > > don't follow postgres development closely, and you have no idea we removed the > > > file. You might work it out from the release notes. But even then, if you try > > > searching for "standby_mode" in the updated docs will tell you basically > > > nothing, it has just vanished > > > > > > Also by simply deleting the page, we've broken web links to https:// > > > www.postgresql.org/docs/current/recovery-config.html > > > > > > So there are three really good reasons: > > > > > > * Help users of the web docs navigate to the right place when we remove things > > > * Don't break web links (breaking links without redirects is bad, the web is > > > sad) > > > * Help upgrading users who know the old terms find the new terms > > > > > > I'd welcome your suggestions on other ways to arrange this, so long as it meets > > > the basic requirement "retain the linktable target 'recovery-config' " > > > > This is certainly not the first or last time we will rename things. > > Indeed, we've renamed things a number of times before, eg: > > https://www.postgresql.org/docs/current/pgwaldump.html > > where the 9.6 link goes to: > > https://www.postgresql.org/docs/9.6/pgxlogdump.html > > and the 'current' link from the 9.6 page goes to the pgwaldump page, > which all works pretty well, if all you're looking at is our > documentation and not considering external links into the documentation. > > However, that isn't what Craig's raising a concern over here (at least, > not exclusively), it's this issue: > > https://www.postgresql.org/docs/current/pgxlogdump.html > > Which currently goes to a 404. > > Now, the pgweb feature that Jonathan wrote recently might actually be > exactly what we need to fix that, and to address the issue with > recovery config documentation that Craig raises. After chatting with Jonathan about this for a bit and testing it out in our test environment, I've gone ahead and added an entry for pgxlogdump.html to redirect to pgwaldump.html, and that seems to be working well. With that then- Craig, can you look at how the pgxlogdump -> pgwaldump pages work and see if using that would address the concerns you've raised here..? Though we need to decide which page 'recovery-config' should go to in newer versions. I'm continuing to chat with Jonathan about if it'd make sense to do the same for the other doc aliases. Thanks, Stephen
Attachment
On Thu, Nov 12, 2020 at 09:47:42AM -0500, Stephen Frost wrote: > > Fortunately, this has already been discussed in the renaming of default > > roles to predefined roles: > > > > https://www.postgresql.org/message-id/flat/157742545062.1149.11052653770497832538%40wrigleys.postgresql.org > > > > This naming change has not happened yet, even though the issue is 11 > > months old, but the agreed-upon way to handle this is to use a website > > redirect that links to the new text. You can add a "tip" there so they > > understand the renaming has happened. > > That rename will suffer the same problem that Craig is concerned about > here regarding the 'current' link, once it's done. I tend to agree with > Craig that it'd be good to improve on this situation, and I've reached > out to Jonathan to ask about using his new feature to have those > /current/ links redirect to the renamed page. I'm actually wondering if > maybe we should just always do that for all the dog page aliases.. > > Might make more sense to discuss this over on -www though. Yes, I am thinking someone could go back and add redirects for previous renames too. It would be interesting also to scrape the web logs for 404 errors to see which renames cause the most failures and do those first. -- Bruce Momjian <bruce@momjian.us> https://momjian.us EnterpriseDB https://enterprisedb.com The usefulness of a cup is in its emptiness, Bruce Lee
On Thu, Nov 12, 2020 at 11:25 PM Stephen Frost <sfrost@snowman.net> wrote:
> Now, the pgweb feature that Jonathan wrote recently might actually be
> exactly what we need to fix that, and to address the issue with
> recovery config documentation that Craig raises.
After chatting with Jonathan about this for a bit and testing it out in
our test environment, I've gone ahead and added an entry for
pgxlogdump.html to redirect to pgwaldump.html, and that seems to be
working well.
Thanks.
With that then- Craig, can you look at how the pgxlogdump -> pgwaldump
pages work and see if using that would address the concerns you've
raised here..?
Though we need to decide which page 'recovery-config' should go to in
newer versions.
Since we basically vanished all evidence of the old configuration, I don't think there is a suitable place.
I maintain that simply vanishing terms from the docs without any sort of explanation is a user-hostile action that we should fix and stop doing If we had something in the docs and we remove it, it's not unduly burdensome to have some index entries that point to the replacement/renamed terms, and a short appendix entry explaining what happened.
If that is for some reason unacceptable (and I don't see anyone giving any actual reason why) the closest I can come up with is probably redirecting to https://www.postgresql.org/docs/current/warm-standby.html#STANDBY-SERVER-OPERATION . But that needs to be fixed to actually explicitly state what makes a standby server into a standby server (per my patch), since right now it just kind of assumes you know about standby.signal .
But... fiddling with the website addresses none of my other concerns. In particular, it doesn't help a user understand that "standby_mode" is gone and to look for "standby.signal" instead. It doesn't provide any "see also" pointers for old terms to point to new terms in the index. Website redirects won't help users with local copies of the docs or manpages who are wondering what the heck happened to recovery.conf and standby_mode either.
So I still think this needs a docs patch. Redirects on the website are not sufficient. If you don't like how I spelled it, consider calling it "important incompatible changes" or something.
The release notes are IMO not sufficient for this because (a) they don't appear in the index; (b) you have to know something has been removed/changed before you know to look in the relnotes for it; (c) you have to find the relnotes for the correct release to find the info you want. An appendix covering important renamings, removals and other incompatible changes would address all those points *and* fix the web links, man page names, etc.
Can anyone tell me why the solution I proposed is not acceptable, and why we have to invent a different one instead? The website redirect is good and all, but doesn't really solve the problem, and I still don't know what's wrong with just fixing the docs...
On Fri, Nov 13, 2020 at 11:31 AM Craig Ringer <craig.ringer@enterprisedb.com> wrote:
Can anyone tell me why the solution I proposed is not acceptable, and why we have to invent a different one instead? The website redirect is good and all, but doesn't really solve the problem, and I still don't know what's wrong with just fixing the docs...
Also, while I'm at it, note that a search on common search engines for "postgres standby" takes you to (an old version of) the hot standby docs. Follow the link to the current docs. Then try to work out from there what exactly makes a server in "archive recovery" or "standby mode".
We need some <link ....> terms on "archive recovery" and "standby mode" there, and probably other places.
I have a draft patch that adds them and various related index cross-referencing in my tree to submit after the recovery.conf docs patch. Let me know if you think that might be worthwhile, 'cos I won't invest time in it if it's going to get reflexively blocked too.
On Fri, Nov 13, 2020 at 11:31:24AM +0800, Craig Ringer wrote: > Can anyone tell me why the solution I proposed is not acceptable, and why we > have to invent a different one instead? The website redirect is good and all, > but doesn't really solve the problem, and I still don't know what's wrong with > just fixing the docs... Because at a certain point the number of _old_ names in the docs obscures exactly how to operate the current software. We have tried keeping stuff around, and we are very bad at removing stuff. -- Bruce Momjian <bruce@momjian.us> https://momjian.us EnterpriseDB https://enterprisedb.com The usefulness of a cup is in its emptiness, Bruce Lee
On Thu, 12 Nov 2020 at 22:31, Craig Ringer <craig.ringer@enterprisedb.com> wrote:
I maintain that simply vanishing terms from the docs without any sort of explanation is a user-hostile action that we should fix and stop doing If we had something in the docs and we remove it, it's not unduly burdensome to have some index entries that point to the replacement/renamed terms, and a short appendix entry explaining what happened.
This sounds very reasonable to me. I would add that while I am by no means an expert in Postgres, although I do know a few things, I will state that it is my professional opinion as a Web person that pages should not simply disappear from formal documentation without some sort of indication of what happened. There are lots of ways to accomplish an indication but for https://www.postgresql.org/docs/current/recovery-config.html or other pages to just disappear is definitely wrong.
On Fri, Nov 13, 2020 at 11:37:16AM +0800, Craig Ringer wrote: > I have a draft patch that adds them and various related index cross-referencing > in my tree to submit after the recovery.conf docs patch. Let me know if you > think that might be worthwhile, 'cos I won't invest time in it if it's going to > get reflexively blocked too. So you are saying you don't think you are getting sufficient thought into your proposal, and getting just a reflex? Just because we don't agree with you don't mean we didn't think about it. In fact, we have thought about it a lot, which is evident from the URL I sent you already. -- Bruce Momjian <bruce@momjian.us> https://momjian.us EnterpriseDB https://enterprisedb.com The usefulness of a cup is in its emptiness, Bruce Lee
On Thu, 12 Nov 2020 at 22:40, Bruce Momjian <bruce@momjian.us> wrote:
Because at a certain point the number of _old_ names in the docs
obscures exactly how to operate the current software. We have tried
keeping stuff around, and we are very bad at removing stuff.
This is a good point, but does not attempt to explain why pages should disappear entirely from /docs/current/. As I said in my previous comment, there are lots of ways of doing this right. For example, we could have pages that would disappear instead be replaced by a short page that explains the page is removed and points to the current documentation of the equivalent or replacement features; these hypothetical "useful 404" (or, more correctly, "useful 410") pages don't even necessarily have to be listed in the table of contents. In fact, serving them with a 410 HTTP status code would be a reasonable thing to do.
On Thu, Nov 12, 2020 at 10:41:49PM -0500, Bruce Momjian wrote: > On Fri, Nov 13, 2020 at 11:37:16AM +0800, Craig Ringer wrote: > > I have a draft patch that adds them and various related index cross-referencing > > in my tree to submit after the recovery.conf docs patch. Let me know if you > > think that might be worthwhile, 'cos I won't invest time in it if it's going to > > get reflexively blocked too. > > So you are saying you don't think you are getting sufficient thought > into your proposal, and getting just a reflex? Just because we don't > agree with you don't mean we didn't think about it. In fact, we have > thought about it a lot, which is evident from the URL I sent you > already. What would be interesting, I think you were suggesting this, is a separate doc chapter that had a list of all the renames, what version they were renamed in, and a link from their to the new name in the docs. This could be easily created by reading the old release notes. Anyone looking for old names would automatically be sent to that page in the docs. This would give us a definitive list, and make the list out of the main flow of the docs. -- Bruce Momjian <bruce@momjian.us> https://momjian.us EnterpriseDB https://enterprisedb.com The usefulness of a cup is in its emptiness, Bruce Lee
On Fri, Nov 13, 2020 at 11:50 AM Bruce Momjian <bruce@momjian.us> wrote:
> So you are saying you don't think you are getting sufficient thought
> into your proposal, and getting just a reflex? Just because we don't
> agree with you don't mean we didn't think about it. In fact, we have
> thought about it a lot, which is evident from the URL I sent you
> already.
I am mostly trying to say that I don't think the issues I raised were actually addressed in the proposed alternatives. I put in a fair bit of effort to clearly set out the problem that this is meant to solve, and was frustrated to perceive the response as "yeah, nah, lets just do this other thing that only addresses one part of the original issue." It wasn't clear why my proposal appeared to be being rejected. Perhaps I didn't fully grasp the context of the linked discussion.
Please review the docs on standbys with a "new user" hat on. It's confusing ( though the new front-matter and definitions in the HA chapter help) even without upgrade considerations. See how long it takes you to determine the answer to the question "what exactly puts a server into 'standby mode' " ?
This proposal was intended to address one part of that, stemming directly from my own challenges with the docs when I as an experienced PostgreSQL user and contributor went to adapt some tests to Pg 12 and 13. I knew we'd removed recovery.conf, but for the life of me I couldn't remember how to put the server in standby mode in 12 or 13 at the time (I've been working with 11 too much lately)... and it took ages to actually find that in the docs.
I can be pretty dense sometimes, but if it sent me for a loop it's going to confuse others a great deal. Two things that would've helped me would've been some cross references to the old configuration terms, and a non-vanishing documentation URL for newer versions. Hence this proposal.
What would be interesting, I think you were suggesting this, is a
separate doc chapter that had a list of all the renames, what version
they were renamed in, and a link from their to the new name in the docs.
Right, that is exactly what I am suggesting we add, as an appendix so it's way out of the way of the main flow of the docs. Per the original patch and the illustrative screenshots. I called it something like "removed and renamed features and settings" or something in the proposed patch. Alternatives would be welcomed, I don't like the name much.
This could be easily created by reading the old release notes. Anyone
looking for old names would automatically be sent to that page in the
docs. This would give us a definitive list, and make the list out of
the main flow of the docs.
On Fri, Nov 13, 2020 at 01:27:34PM +0800, Craig Ringer wrote:
> On Fri, Nov 13, 2020 at 11:50 AM Bruce Momjian <bruce@momjian.us> wrote:
>
>
> > So you are saying you don't think you are getting sufficient thought
> > into your proposal, and getting just a reflex? Just because we don't
> > agree with you don't mean we didn't think about it. In fact, we have
> > thought about it a lot, which is evident from the URL I sent you
> > already.
>
>
> I am mostly trying to say that I don't think the issues I raised were actually
> addressed in the proposed alternatives. I put in a fair bit of effort to
> clearly set out the problem that this is meant to solve, and was frustrated to
> perceive the response as "yeah, nah, lets just do this other thing that only
> addresses one part of the original issue." It wasn't clear why my proposal
> appeared to be being rejected. Perhaps I didn't fully grasp the context of the
> linked discussion.
I think the big problem, and I have seen this repeatedly, is showing up
with a patch without discussing whether people actually want the
feature. I know it is a doc issue, but our TODO list has the order as:
Desirability -> Design -> Implement -> Test -> Review -> Commit
and there is a reason for that. When you appear with a patch, you are
already far down the steps, and you have to back up to explain why it is
useful.
Clearly we have need for documenting these renamings somewhere. We were
going to go with a simple URL redirect and a "tip" for
default/pre-installed roles, but I like the idea of doing something more
wholistic that covers all of our recent renaming cases. Let's get
buy-in from that, and then someone can work on a patch.
--
Bruce Momjian <bruce@momjian.us> https://momjian.us
EnterpriseDB https://enterprisedb.com
The usefulness of a cup is in its emptiness, Bruce Lee
On Fri, Nov 13, 2020 at 10:42 AM Bruce Momjian <bruce@momjian.us> wrote:
I think the big problem, and I have seen this repeatedly, is showing up
with a patch without discussing whether people actually want the
feature. I know it is a doc issue, but our TODO list has the order as:
Desirability -> Design -> Implement -> Test -> Review -> Commit
and there is a reason for that. When you appear with a patch, you are
already far down the steps, and you have to back up to explain why it is
useful.
That process is designed to prevent people from being exposed to wasted effort and hard feelings. The choice to follow it individually, instead of collectively, doesn't diminish the value of the end result.
I generally agree with Craig's proposed solution here. It doesn't add any cognitive load to new users as they will not see the obsolete features appendix in the normal course of their reading.
To the particular point regarding renaming features - this situation is not an instance of a rename but rather a feature removal. To blindly apply the reasoning and decision made for renaming to removal is not reasonable. From that observation (and the commentary below) extends the conclusion that this appendix shouldn't include renaming.
On the point of renaming, my suggestion would be to have the documentation directory provide a file of all renaming for which redirects should be performed. pgweb would source that file and actually establish the redirects on the main website. Comments in the file can describe to a curious user why the name change was needed. Though that honestly seems a bit overkill; for rename, the content as a whole still exists and a comment therein can talk about the renaming. Users of the public website would still get the benefit of redirects, and there isn't any practical reason for people building documentation from source to want to establish such redirects even if they were provided the data in the form of the aforementioned file.
I believe there is probably room for more discussion regarding the value of providing a limited view of history in the publicly facing documentation but that seems outside the scope of this patch.
David J.
On Sun, Nov 15, 2020 at 1:49 PM David G. Johnston <david.g.johnston@gmail.com> wrote:
On Fri, Nov 13, 2020 at 10:42 AM Bruce Momjian <bruce@momjian.us> wrote:I think the big problem, and I have seen this repeatedly, is showing up
with a patch without discussing whether people actually want the
feature. I know it is a doc issue, but our TODO list has the order as:
Desirability -> Design -> Implement -> Test -> Review -> Commit
and there is a reason for that. When you appear with a patch, you are
already far down the steps, and you have to back up to explain why it is
useful.That process is designed to prevent people from being exposed to wasted effort and hard feelings. The choice to follow it individually, instead of collectively, doesn't diminish the value of the end result.
Frankly, it's also kind of a catch-22. Because often proposals for changes or discussion get ignored until there's a patch, or the response is along the lines of "show us a patch so we can try it and get a solid idea of what this will do."
For major engineering changes, yes, discuss first. For small stuff, if you don't want to get ignored, open with a patch.
On the point of renaming, my suggestion would be to have the documentation directory provide a file of all renaming for which redirects should be performed. pgweb would source that file and actually establish the redirects on the main website. Comments in the file can describe to a curious user why the name change was needed. Though that honestly seems a bit overkill; for rename, the content as a whole still exists and a comment therein can talk about the renaming. Users of the public website would still get the benefit of redirects, and there isn't any practical reason for people building documentation from source to want to establish such redirects even if they were provided the data in the form of the aforementioned file.
Agreed, there's no need to keep heading redirects in the source-built docs. So if we're happy to maintain that on the website, in a way that makes /current/ links work *and* following links from a /11/ docs page that has vanished in pg12 to the right new place via the version navigation links, that's sufficient for topic-heading renames.
We should, however, carry information about removals and renames in the source-built docs to the extent that we have appropriate "see also" index entries and useful information somewhere in the docs for people who are upgrading. It just doesn't have to retain the same subject heading.
On Sat, Nov 14, 2020 at 1:42 AM Bruce Momjian <bruce@momjian.us> wrote:
Clearly we have need for documenting these renamings somewhere. We were
going to go with a simple URL redirect and a "tip" for
default/pre-installed roles, but I like the idea of doing something more
wholistic that covers all of our recent renaming cases. Let's get
buy-in from that, and then someone can work on a patch.
Is there anything further I can do to address this specific documentation issue?
Can I get you to consider the user experience arising from the current docs - try using the docs to find out how to set up a standby?
I'm not prepared to expand the scope of this to do a major review of all past renamings and changes without a very clear agreement about how that should be implemented in the docs and how that will address all the relevant UX issues - vanishing terms from indexes and docs without x-refs/see-alsos, vanishing URLs endpoints for public links, vanishing next-version links in www docs.
I didn't raise this for discussion before I submitted a patch because I thought it was such an obvious no-brainer that a simple patch to address an obviously confusing aspect of the docs after the recovery.conf removal would be uncontroversial. Anyway, as I've noted, these things often get ignored until there is a patch to argue about.
Can we please just address this docs issue? If you don't like my solution can you please supply a patch that you feel addresses the problem? Or clearly state that you don't think there is a problem, and do so in a way that actually addresses the specific points I have raised about what's wrong with the status quo?
Greetings, * Craig Ringer (craig.ringer@enterprisedb.com) wrote: > On Sat, Nov 14, 2020 at 1:42 AM Bruce Momjian <bruce@momjian.us> wrote: > > Clearly we have need for documenting these renamings somewhere. We were > > going to go with a simple URL redirect and a "tip" for > > default/pre-installed roles, but I like the idea of doing something more > > wholistic that covers all of our recent renaming cases. Let's get > > buy-in from that, and then someone can work on a patch. > > Is there anything further I can do to address this specific documentation > issue? > > Can I get you to consider the user experience arising from the current docs > - try using the docs to find out how to set up a standby? For my 2c, at least, I'm in favor of making some kind of change here to make things better for users. I tried to figure out a way using the features we have easily available in pgweb, but I tend to agree with Craig that those just aren't enough for the more recent set of changes that we've made. > I'm not prepared to expand the scope of this to do a major review of all > past renamings and changes without a very clear agreement about how that > should be implemented in the docs and how that will address all the > relevant UX issues - vanishing terms from indexes and docs without > x-refs/see-alsos, vanishing URLs endpoints for public links, vanishing > next-version links in www docs. The past renamings haven't really been as much of an issue since the redirects and such, imv anyway, have been sufficient to deal with them. > Can we please just address this docs issue? If you don't like my solution > can you please supply a patch that you feel addresses the problem? Or > clearly state that you don't think there is a problem, and do so in a way > that actually addresses the specific points I have raised about what's > wrong with the status quo? While I understand not wanting to go back through and check for all renamings, it seems like we should probably at least list out the ones we know about pretty easily, if we're going to create a new section specifically for those cases..? Or do we think the current approach works well enough for those other cases but just not for this one? Think I listed this elsewhere but not seeing it on the thread so I'll include it here anyway. These are the current doc aliases: catalog-pg-replication-slots.html <-> view-pg-replication-slots.html pgxlogdump.html <-> pgwaldump.html app-pgresetxlog.html <-> app-pgresetwal.html legalnotice.html <-> LEGALNOTICE.html app-pgreceivexlog.html <-> app-pgreceivewal.html As relates to the specific patch, I don't think the comments line up quite right (we can prevent a 404 from happening through other means, but the point of the patch is really to give a deeper explanation of what happened). Also- wrt pg_basebackup, isn't that expected to support older versions, and so we should document the behavior against older versions..? Thanks, Stephen
Attachment
On Mon, Nov 30, 2020 at 10:11:04AM +0800, Craig Ringer wrote: > Can we please just address this docs issue? If you don't like my solution can > you please supply a patch that you feel addresses the problem? Or clearly state > that you don't think there is a problem, and do so in a way that actually > addresses the specific points I have raised about what's wrong with the status > quo? If we know there are X problems, and we fix one of them one way, then later fix the rest another way, we have to undo the first fix. If you don't want to fix all X, then let's wait until someone does want to fix them all. -- Bruce Momjian <bruce@momjian.us> https://momjian.us EnterpriseDB https://enterprisedb.com The usefulness of a cup is in its emptiness, Bruce Lee
On Mon, Nov 30, 2020 at 11:25 AM Bruce Momjian <bruce@momjian.us> wrote:
On Mon, Nov 30, 2020 at 10:11:04AM +0800, Craig Ringer wrote:
> Can we please just address this docs issue? If you don't like my solution can
> you please supply a patch that you feel addresses the problem? Or clearly state
> that you don't think there is a problem, and do so in a way that actually
> addresses the specific points I have raised about what's wrong with the status
> quo?
If we know there are X problems, and we fix one of them one way, then
later fix the rest another way, we have to undo the first fix. If you
don't want to fix all X, then let's wait until someone does want to fix
them all.
IMO there is only the original problem with an acceptable solution presented that can be committed without downside. If that has to be undone because someone else in the future decides on a different solution that happens to touch this too, fine, it can be changed again.
David J.
On Mon, Nov 30, 2020 at 11:31:35AM -0700, David G. Johnston wrote: > On Mon, Nov 30, 2020 at 11:25 AM Bruce Momjian <bruce@momjian.us> wrote: > > On Mon, Nov 30, 2020 at 10:11:04AM +0800, Craig Ringer wrote: > > Can we please just address this docs issue? If you don't like my solution > can > > you please supply a patch that you feel addresses the problem? Or clearly > state > > that you don't think there is a problem, and do so in a way that actually > > addresses the specific points I have raised about what's wrong with the > status > > quo? > > If we know there are X problems, and we fix one of them one way, then > later fix the rest another way, we have to undo the first fix. If you > don't want to fix all X, then let's wait until someone does want to fix > them all. > > IMO there is only the original problem with an acceptable solution presented > that can be committed without downside. If that has to be undone because > someone else in the future decides on a different solution that happens to > touch this too, fine, it can be changed again. The downside is you end up with X-1 dummy sections just to allow for references to old syntax, and you then have to find them all and remove them when you implement the proper solution. I have no intention of applying such an X-1 fix. -- Bruce Momjian <bruce@momjian.us> https://momjian.us EnterpriseDB https://enterprisedb.com The usefulness of a cup is in its emptiness, Bruce Lee
On Mon, Nov 30, 2020 at 11:42 AM Bruce Momjian <bruce@momjian.us> wrote:
The downside is you end up with X-1 dummy sections just to allow for
references to old syntax, and you then have to find them all and remove
them when you implement the proper solution. I have no intention of
applying such an X-1 fix.
X = 2; seems like a strong objection for such a minor issue. The status quo seems worse than that.
David J.
Greetings, * David G. Johnston (david.g.johnston@gmail.com) wrote: > On Mon, Nov 30, 2020 at 11:42 AM Bruce Momjian <bruce@momjian.us> wrote: > > The downside is you end up with X-1 dummy sections just to allow for > > references to old syntax, and you then have to find them all and remove > > them when you implement the proper solution. I have no intention of > > applying such an X-1 fix. > > X = 2; seems like a strong objection for such a minor issue. The status > quo seems worse than that. I've been thinking about this and I think I'm on Craig and David's side- having something cleaner, and clearer, than just http redirects and such would be good for these cases and I don't think we are going to end up with so many of them that it ends up becoming an issue. Thanks, Stephen
Attachment
On Wed, Dec 2, 2020 at 02:47:13PM -0500, Stephen Frost wrote: > Greetings, > > * David G. Johnston (david.g.johnston@gmail.com) wrote: > > On Mon, Nov 30, 2020 at 11:42 AM Bruce Momjian <bruce@momjian.us> wrote: > > > The downside is you end up with X-1 dummy sections just to allow for > > > references to old syntax, and you then have to find them all and remove > > > them when you implement the proper solution. I have no intention of > > > applying such an X-1 fix. > > > > X = 2; seems like a strong objection for such a minor issue. The status > > quo seems worse than that. > > I've been thinking about this and I think I'm on Craig and David's side- > having something cleaner, and clearer, than just http redirects and such > would be good for these cases and I don't think we are going to end up > with so many of them that it ends up becoming an issue. We were not going to use just redirects --- we were going to create a page that had all the renames listed, with links to the new names. -- Bruce Momjian <bruce@momjian.us> https://momjian.us EnterpriseDB https://enterprisedb.com The usefulness of a cup is in its emptiness, Bruce Lee
Greetings, * Bruce Momjian (bruce@momjian.us) wrote: > On Wed, Dec 2, 2020 at 02:47:13PM -0500, Stephen Frost wrote: > > * David G. Johnston (david.g.johnston@gmail.com) wrote: > > > On Mon, Nov 30, 2020 at 11:42 AM Bruce Momjian <bruce@momjian.us> wrote: > > > > The downside is you end up with X-1 dummy sections just to allow for > > > > references to old syntax, and you then have to find them all and remove > > > > them when you implement the proper solution. I have no intention of > > > > applying such an X-1 fix. > > > > > > X = 2; seems like a strong objection for such a minor issue. The status > > > quo seems worse than that. > > > > I've been thinking about this and I think I'm on Craig and David's side- > > having something cleaner, and clearer, than just http redirects and such > > would be good for these cases and I don't think we are going to end up > > with so many of them that it ends up becoming an issue. > > We were not going to use just redirects --- we were going to create a > page that had all the renames listed, with links to the new names. Maybe I'm the one who is confused here, but I thought there was objection to adding a new section/page which covers these topics (which is what Craig's original patch does)...? If there isn't an objection to that then it seems like we should move forward with it. If I'm following correctly, maybe there was some idea that we should have more things added to this section than just the recovery.conf bits, and perhaps we should, but that could certainly be done later. To be clear though, I don't think we need to do this in all cases- the existing flow for pg_xlogdump -> pg_waldump works pretty well. Maybe we add in a note here too if someone wants to but I don't think it's strictly necessary for the 'simple' rename cases. I also feel like that could be done once the section gets added, if someone wants to. Was there something else that I'm missing here in terms of what the concern is regarding Craig's patch..? Thanks, Stephen
Attachment
On Wed, Dec 2, 2020 at 05:57:01PM -0500, Stephen Frost wrote: > * Bruce Momjian (bruce@momjian.us) wrote: > > We were not going to use just redirects --- we were going to create a > > page that had all the renames listed, with links to the new names. > > Maybe I'm the one who is confused here, but I thought there was > objection to adding a new section/page which covers these topics (which > is what Craig's original patch does)...? If there isn't an objection to > that then it seems like we should move forward with it. > > If I'm following correctly, maybe there was some idea that we should > have more things added to this section than just the recovery.conf bits, > and perhaps we should, but that could certainly be done later. To be > clear though, I don't think we need to do this in all cases- the > existing flow for pg_xlogdump -> pg_waldump works pretty well. Maybe we > add in a note here too if someone wants to but I don't think it's > strictly necessary for the 'simple' rename cases. > > I also feel like that could be done once the section gets added, if > someone wants to. > > Was there something else that I'm missing here in terms of what the > concern is regarding Craig's patch..? I think the ideal solution is to create a section for all the rename cases and do all the redirects to that page. The page would list the old and new name for each item, and would link to the section for each new item. -- Bruce Momjian <bruce@momjian.us> https://momjian.us EnterpriseDB https://enterprisedb.com The usefulness of a cup is in its emptiness, Bruce Lee
On Wed, Dec 2, 2020 at 5:26 PM Bruce Momjian <bruce@momjian.us> wrote:
I think the ideal solution is to create a section for all the rename
cases and do all the redirects to that page. The page would list the
old and new name for each item, and would link to the section for each
new item.
Nothing prevents us from doing that for simple renames. For me, this situation is not a simple rename and the proposed solution is appropriate for what it is - changing the implementation details of an existing feature. We can do both - though the simple rename page doesn't seem particularly appealing at first glance.
David J.
On Wed, 2 Dec 2020 at 19:33, David G. Johnston <david.g.johnston@gmail.com> wrote:
On Wed, Dec 2, 2020 at 5:26 PM Bruce Momjian <bruce@momjian.us> wrote:I think the ideal solution is to create a section for all the rename
cases and do all the redirects to that page. The page would list the
old and new name for each item, and would link to the section for each
new item.Nothing prevents us from doing that for simple renames. For me, this situation is not a simple rename and the proposed solution is appropriate for what it is - changing the implementation details of an existing feature. We can do both - though the simple rename page doesn't seem particularly appealing at first glance.
I for one do not like following a bookmark or link and then being redirected to a generic page that doesn't relate to the specific link I was following. What is being proposed here is not as bad as the usual, where all the old links simply turn into redirects to the homepage, but it's still disorienting. I would much rather each removed page be moved to an appendix (without renaming) and edited to briefly explain what happened to the page and provide links to the appropriate up-to-date page or pages.
On Wed, Dec 2, 2020 at 08:07:47PM -0500, Isaac Morland wrote: > On Wed, 2 Dec 2020 at 19:33, David G. Johnston <david.g.johnston@gmail.com> > wrote: > > On Wed, Dec 2, 2020 at 5:26 PM Bruce Momjian <bruce@momjian.us> wrote: > > I think the ideal solution is to create a section for all the rename > cases and do all the redirects to that page. The page would list the > old and new name for each item, and would link to the section for each > new item. > > > > Nothing prevents us from doing that for simple renames. For me, this > situation is not a simple rename and the proposed solution is appropriate > for what it is - changing the implementation details of an existing > feature. We can do both - though the simple rename page doesn't seem > particularly appealing at first glance. > > > I for one do not like following a bookmark or link and then being redirected to > a generic page that doesn't relate to the specific link I was following. What > is being proposed here is not as bad as the usual, where all the old links > simply turn into redirects to the homepage, but it's still disorienting. I > would much rather each removed page be moved to an appendix (without renaming) > and edited to briefly explain what happened to the page and provide links to > the appropriate up-to-date page or pages. Yes, that is pretty much the same thing I was suggesting, except that each rename has its own _original_ URL link, which I think is also acceptable. My desire is for these items to all exist in one place, and an appendix of them seems fine. -- Bruce Momjian <bruce@momjian.us> https://momjian.us EnterpriseDB https://enterprisedb.com The usefulness of a cup is in its emptiness, Bruce Lee
Greetings, * Bruce Momjian (bruce@momjian.us) wrote: > On Wed, Dec 2, 2020 at 08:07:47PM -0500, Isaac Morland wrote: > > On Wed, 2 Dec 2020 at 19:33, David G. Johnston <david.g.johnston@gmail.com> > > wrote: > > > > On Wed, Dec 2, 2020 at 5:26 PM Bruce Momjian <bruce@momjian.us> wrote: > > > > I think the ideal solution is to create a section for all the rename > > cases and do all the redirects to that page. The page would list the > > old and new name for each item, and would link to the section for each > > new item. > > > > > > > > Nothing prevents us from doing that for simple renames. For me, this > > situation is not a simple rename and the proposed solution is appropriate > > for what it is - changing the implementation details of an existing > > feature. We can do both - though the simple rename page doesn't seem > > particularly appealing at first glance. > > > > > > I for one do not like following a bookmark or link and then being redirected to > > a generic page that doesn't relate to the specific link I was following. What > > is being proposed here is not as bad as the usual, where all the old links > > simply turn into redirects to the homepage, but it's still disorienting. I > > would much rather each removed page be moved to an appendix (without renaming) > > and edited to briefly explain what happened to the page and provide links to > > the appropriate up-to-date page or pages. > > Yes, that is pretty much the same thing I was suggesting, except that > each rename has its own _original_ URL link, which I think is also > acceptable. My desire is for these items to all exist in one place, and > an appendix of them seems fine. Alright, so, to try and move this forward I'll list out (again) the renames that we have in pgweb: catalog-pg-replication-slots.html <-> view-pg-replication-slots.html pgxlogdump.html <-> pgwaldump.html app-pgresetxlog.html <-> app-pgresetwal.html app-pgreceivexlog.html <-> app-pgreceivewal.html (excluding the 'legal notice' one) Bruce, are you saying that we need to take Craig's patch and then add to it entries for all of the above, effectively removing the need for the web page aliases and redirects? If that was done, would that be sufficient to get this committed? Are there other things that people can think of off-hand that we should include, I think Craig might have mentioned something else earlier on..? I don't think we should require that someone troll through everything that ever existed, just to be clear, as we can always add to this later if other things come up. If that's the expectation though, then someone needs to say so, in which case I'll assume it's status quo unless/until someone steps up to do that. Obviously, I'd then have to adjust the patch that I proposed for default roles, or move forward with it as-is, depending on what we end up doing here. I dislike what feels like a state of limbo for this right now though. Thanks, Stephen
Attachment
On Fri, Dec 4, 2020 at 12:00 PM Stephen Frost <sfrost@snowman.net> wrote:
Obviously, I'd then have to adjust the patch that I proposed for default
roles, or move forward with it as-is, depending on what we end up doing
here. I dislike what feels like a state of limbo for this right now
though.
We have a committer + 1 in favor of status quo - or at least requiring that a non-existing solution be created to move things forward. (Bruce, Daniel)
We have a committer + 3 that seem to agree that the proposed patch is acceptable as presented. (Stephen, Craig, Isaac, David J.)
Anyone wish to update the above observation?
Stephen, are you will to commit this with the support of the community members who have spoken up here?
David J.
Greetings, * David G. Johnston (david.g.johnston@gmail.com) wrote: > On Fri, Dec 4, 2020 at 12:00 PM Stephen Frost <sfrost@snowman.net> wrote: > > Obviously, I'd then have to adjust the patch that I proposed for default > > roles, or move forward with it as-is, depending on what we end up doing > > here. I dislike what feels like a state of limbo for this right now > > though. > > We have a committer + 1 in favor of status quo - or at least requiring that > a non-existing solution be created to move things forward. (Bruce, Daniel) > We have a committer + 3 that seem to agree that the proposed patch is > acceptable as presented. (Stephen, Craig, Isaac, David J.) > Anyone wish to update the above observation? > > Stephen, are you will to commit this with the support of the community > members who have spoken up here? What I was hoping to achieve is consensus on a reasonably well bounded solution. I'm not going to push something that others are objecting to, but if we can agree on the idea and what's needed for it to be acceptable then I'm willing to work towards that, provided it doesn't require going back 10 versions and looking at every single change. Thanks, Stephen
Attachment
On Fri, Dec 4, 2020 at 02:00:23PM -0500, Stephen Frost wrote: > * Bruce Momjian (bruce@momjian.us) wrote: > > Yes, that is pretty much the same thing I was suggesting, except that > > each rename has its own _original_ URL link, which I think is also > > acceptable. My desire is for these items to all exist in one place, and > > an appendix of them seems fine. > > Alright, so, to try and move this forward I'll list out (again) the > renames that we have in pgweb: > > catalog-pg-replication-slots.html <-> view-pg-replication-slots.html > pgxlogdump.html <-> pgwaldump.html > app-pgresetxlog.html <-> app-pgresetwal.html > app-pgreceivexlog.html <-> app-pgreceivewal.html > > (excluding the 'legal notice' one) > > Bruce, are you saying that we need to take Craig's patch and then add to > it entries for all of the above, effectively removing the need for the > web page aliases and redirects? If that was done, would that be Yes, I think putting the compatibility section headings in our main documentation flow will make it too hard to read and cause unnecessary complexity, but if we have a separate section for them, adding the section headings seems fine. This way, we don't have to add a redirect every time we add a new entry. > sufficient to get this committed? Are there other things that people > can think of off-hand that we should include, I think Craig might have > mentioned something else earlier on..? I don't think we should require > that someone troll through everything that ever existed, just to be > clear, as we can always add to this later if other things come up. If > that's the expectation though, then someone needs to say so, in which > case I'll assume it's status quo unless/until someone steps up to do > that. Agreed. I just wanted something that could scale going forward, and be easily identified as compatibility, so maybe one day we can remove them. However, if they are in a separate section, we might never do that. -- Bruce Momjian <bruce@momjian.us> https://momjian.us EnterpriseDB https://enterprisedb.com The usefulness of a cup is in its emptiness, Bruce Lee
Greetings, * Bruce Momjian (bruce@momjian.us) wrote: > On Fri, Dec 4, 2020 at 02:00:23PM -0500, Stephen Frost wrote: > > * Bruce Momjian (bruce@momjian.us) wrote: > > > Yes, that is pretty much the same thing I was suggesting, except that > > > each rename has its own _original_ URL link, which I think is also > > > acceptable. My desire is for these items to all exist in one place, and > > > an appendix of them seems fine. > > > > Alright, so, to try and move this forward I'll list out (again) the > > renames that we have in pgweb: > > > > catalog-pg-replication-slots.html <-> view-pg-replication-slots.html > > pgxlogdump.html <-> pgwaldump.html > > app-pgresetxlog.html <-> app-pgresetwal.html > > app-pgreceivexlog.html <-> app-pgreceivewal.html > > > > (excluding the 'legal notice' one) > > > > Bruce, are you saying that we need to take Craig's patch and then add to > > it entries for all of the above, effectively removing the need for the > > web page aliases and redirects? If that was done, would that be > > Yes, I think putting the compatibility section headings in our main > documentation flow will make it too hard to read and cause unnecessary > complexity, but if we have a separate section for them, adding the > section headings seems fine. This way, we don't have to add a redirect > every time we add a new entry. Alright, how does this look? The new entries are all under the 'obsolete' section to keep it out of the main line, but should work to 'fix' the links that currently 404 and provide a bit of a 'softer' landing for the other cases that currently just forcibly redirect using the website doc alias capability. I ended up not actually doing this for the catalog -> view change of pg_replication_slots simply because I don't really think folks will misunderstand or be confused by that redirect since it's still the same relation. If others disagree though, we could certainly change that too. > > sufficient to get this committed? Are there other things that people > > can think of off-hand that we should include, I think Craig might have > > mentioned something else earlier on..? I don't think we should require > > that someone troll through everything that ever existed, just to be > > clear, as we can always add to this later if other things come up. If > > that's the expectation though, then someone needs to say so, in which > > case I'll assume it's status quo unless/until someone steps up to do > > that. > > Agreed. I just wanted something that could scale going forward, and be > easily identified as compatibility, so maybe one day we can remove them. > However, if they are in a separate section, we might never do that. Sure, seems like this approach addresses that. If we have agreement from folks on this then I'll commit it and then rework the change from default roles to predefined roles to use this approach and then we can move forward with that too. Thanks, Stephen
Attachment
On Thu, 14 Jan 2021 at 03:44, Stephen Frost <sfrost@snowman.net> wrote:
Alright, how does this look? The new entries are all under the
'obsolete' section to keep it out of the main line, but should work to
'fix' the links that currently 404 and provide a bit of a 'softer'
landing for the other cases that currently just forcibly redirect using
the website doc alias capability.
Thanks for expanding the change to other high profile obsoleted or renamed features and tools.
One minor point. I'm not sure this is quite the best way to spell the index entries:
+ <indexterm>
+ <primary>obsolete</primary>
+ <secondary>pg_receivexlog</secondary>
+ </indexterm>
as it will produce an index term "obsolete" with a list of various components under it. While that concentrates them nicely, it means people won't actually find them if they're using the index alphabetically.
I'd slightly prefer
+ <indexterm>
+ <primary>pg_receivexlog</primary>
+ <seealso>pg_receivewal</secondary>
+ </indexterm>
even though that bulks the index up a little, because then people are a bit more likely to find it.
Your extended and revised patch retains the above style for
+ <indexterm>
+ <primary>trigger_file</primary>
+ <seealso>promote_trigger_file</seealso>
+ </indexterm>
+ <primary>trigger_file</primary>
+ <seealso>promote_trigger_file</seealso>
+ </indexterm>
...
+ <indexterm>
+ <primary>standby_mode</primary>
+ <seealso>standby.signal</seealso>
+ </indexterm>
+ <primary>standby_mode</primary>
+ <seealso>standby.signal</seealso>
+ </indexterm>
so if you intend to change it, that entry needs changing too.
I ended up not actually doing this for the catalog -> view change of
pg_replication_slots simply because I don't really think folks will
misunderstand or be confused by that redirect since it's still the same
relation. If others disagree though, we could certainly change that
too.
I agree with you.
Greetings, * Craig Ringer (craig.ringer@enterprisedb.com) wrote: > On Thu, 14 Jan 2021 at 03:44, Stephen Frost <sfrost@snowman.net> wrote: > > Alright, how does this look? The new entries are all under the > > 'obsolete' section to keep it out of the main line, but should work to > > 'fix' the links that currently 404 and provide a bit of a 'softer' > > landing for the other cases that currently just forcibly redirect using > > the website doc alias capability. > > Thanks for expanding the change to other high profile obsoleted or renamed > features and tools. Thanks for taking the time to review it and comment on it! > One minor point. I'm not sure this is quite the best way to spell the index > entries: > > + <indexterm> > + <primary>obsolete</primary> > + <secondary>pg_receivexlog</secondary> > + </indexterm> > > as it will produce an index term "obsolete" with a list of various > components under it. While that concentrates them nicely, it means people > won't actually find them if they're using the index alphabetically. Ah, yeah, that's definitely a good point and one that I hadn't really spent much time thinking about. > I'd slightly prefer > > + <indexterm> > + <primary>pg_receivexlog</primary> > + <seealso>pg_receivewal</secondary> > + </indexterm> > > even though that bulks the index up a little, because then people are a bit > more likely to find it. Yup, makes sense, updated patch attached which makes that change. > > I ended up not actually doing this for the catalog -> view change of > > pg_replication_slots simply because I don't really think folks will > > misunderstand or be confused by that redirect since it's still the same > > relation. If others disagree though, we could certainly change that > > too. > > I agree with you. Ok, great. How does the attached look then? Bruce, did you want to review or comment on this as to if it addresses your concerns appropriately? Would be great to get this in as there's the follow-on for default roles. Thanks! Stephen
Attachment
Greetings, * Stephen Frost (sfrost@snowman.net) wrote: > * Craig Ringer (craig.ringer@enterprisedb.com) wrote: > > On Thu, 14 Jan 2021 at 03:44, Stephen Frost <sfrost@snowman.net> wrote: > > > Alright, how does this look? The new entries are all under the > > > 'obsolete' section to keep it out of the main line, but should work to > > > 'fix' the links that currently 404 and provide a bit of a 'softer' > > > landing for the other cases that currently just forcibly redirect using > > > the website doc alias capability. > > > > Thanks for expanding the change to other high profile obsoleted or renamed > > features and tools. > > Thanks for taking the time to review it and comment on it! > > > One minor point. I'm not sure this is quite the best way to spell the index > > entries: > > > > + <indexterm> > > + <primary>obsolete</primary> > > + <secondary>pg_receivexlog</secondary> > > + </indexterm> > > > > as it will produce an index term "obsolete" with a list of various > > components under it. While that concentrates them nicely, it means people > > won't actually find them if they're using the index alphabetically. > > Ah, yeah, that's definitely a good point and one that I hadn't really > spent much time thinking about. > > > I'd slightly prefer > > > > + <indexterm> > > + <primary>pg_receivexlog</primary> > > + <seealso>pg_receivewal</secondary> > > + </indexterm> > > > > even though that bulks the index up a little, because then people are a bit > > more likely to find it. > > Yup, makes sense, updated patch attached which makes that change. > > > > I ended up not actually doing this for the catalog -> view change of > > > pg_replication_slots simply because I don't really think folks will > > > misunderstand or be confused by that redirect since it's still the same > > > relation. If others disagree though, we could certainly change that > > > too. > > > > I agree with you. > > Ok, great. > > How does the attached look then? > > Bruce, did you want to review or comment on this as to if it addresses > your concerns appropriately? Would be great to get this in as there's > the follow-on for default roles. ... really attached now, sorry about that. :) Thanks, Stephen
Attachment
On Wed, 20 Jan 2021 at 02:45, Stephen Frost <sfrost@snowman.net> wrote:
Greetings,
* Stephen Frost (sfrost@snowman.net) wrote:
> * Craig Ringer (craig.ringer@enterprisedb.com) wrote:
> > On Thu, 14 Jan 2021 at 03:44, Stephen Frost <sfrost@snowman.net> wrote:
> > > Alright, how does this look? The new entries are all under the
> > > 'obsolete' section to keep it out of the main line, but should work to
> > > 'fix' the links that currently 404 and provide a bit of a 'softer'
> > > landing for the other cases that currently just forcibly redirect using
> > > the website doc alias capability.
> >
> > Thanks for expanding the change to other high profile obsoleted or renamed
> > features and tools.
>
> Thanks for taking the time to review it and comment on it!
>
> > One minor point. I'm not sure this is quite the best way to spell the index
> > entries:
> >
> > + <indexterm>
> > + <primary>obsolete</primary>
> > + <secondary>pg_receivexlog</secondary>
> > + </indexterm>
> >
> > as it will produce an index term "obsolete" with a list of various
> > components under it. While that concentrates them nicely, it means people
> > won't actually find them if they're using the index alphabetically.
>
> Ah, yeah, that's definitely a good point and one that I hadn't really
> spent much time thinking about.
>
> > I'd slightly prefer
> >
> > + <indexterm>
> > + <primary>pg_receivexlog</primary>
> > + <seealso>pg_receivewal</secondary>
> > + </indexterm>
> >
> > even though that bulks the index up a little, because then people are a bit
> > more likely to find it.
>
> Yup, makes sense, updated patch attached which makes that change.
>
> > > I ended up not actually doing this for the catalog -> view change of
> > > pg_replication_slots simply because I don't really think folks will
> > > misunderstand or be confused by that redirect since it's still the same
> > > relation. If others disagree though, we could certainly change that
> > > too.
> >
> > I agree with you.
>
> Ok, great.
>
> How does the attached look then?
Pretty good to me. Thanks so much for your help and support with this.
Index entries render as e.g.
pg_xlogdump, The pg_xlogdump command
(see also pg_waldump)
(see also pg_waldump)
wheras with the obsolete subhead they would render as something like:
obsolete, Obsolete or renamed features, settings and files
pg_xlogdump, The pg_xlogdump command
The see also spelling is much easier to find in the index but doesn't make it as obvious that it's obsoleted/replaced.
A look at the doxygen docs suggest we should use <see> not <seealso> for these.
A quick
sed -i -e 's/<seealso>/<see>/g' -e 's/<\/seealso>/<\/see>/g' doc/src/sgml/appendix-obsolete*
causes them to render much better:
pg_receivexlog, The pg_receivexlog command (see pg_receivewal)
It might be worth changing the <title/>s too, so I've done so in the attached. The terms now render as:
pg_receivexlog, pg_receivexlog renamed to pg_recievewal (see pg_receivewal)
which is good enough in my opinion. The duplication is messy but an expected artifact of index generation. I don't see any docbook <indexterm> attribute that lets you suppress insertion of the <title> of the section containing the <indexterm>, and it's not worth fiddling to try to eliminate it with structural hacks.
The attached changes the titles, changes <seealso> to <see>, and also updates the comments in the obsolete entries SGML docs to specify that the id must be unchanged + give a recommended index term format.
Attachment
Greetings, * Craig Ringer (craig.ringer@enterprisedb.com) wrote: > Pretty good to me. Thanks so much for your help and support with this. Thanks for helping me move it forward! > Index entries render as e.g. > > pg_xlogdump, The pg_xlogdump command > (see also pg_waldump) > > wheras with the obsolete subhead they would render as something like: > > obsolete, Obsolete or renamed features, settings and files > pg_xlogdump, The pg_xlogdump command > > The see also spelling is much easier to find in the index but doesn't make > it as obvious that it's obsoleted/replaced. > > A look at the doxygen docs suggest we should use <see> not <seealso> for > these. > > A quick > > sed -i -e 's/<seealso>/<see>/g' -e 's/<\/seealso>/<\/see>/g' > doc/src/sgml/appendix-obsolete* > > causes them to render much better: > > pg_receivexlog, The pg_receivexlog command (see pg_receivewal) > > It might be worth changing the <title/>s too, so I've done so in the > attached. The terms now render as: > > pg_receivexlog, pg_receivexlog renamed to pg_recievewal (see > pg_receivewal) > > which is good enough in my opinion. The duplication is messy but an > expected artifact of index generation. I don't see any docbook <indexterm> > attribute that lets you suppress insertion of the <title> of the section > containing the <indexterm>, and it's not worth fiddling to try to eliminate > it with structural hacks. Nice, yes, that does look better. > The attached changes the titles, changes <seealso> to <see>, and also > updates the comments in the obsolete entries SGML docs to specify that the > id must be unchanged + give a recommended index term format. Awesome, attached is just a rebase (not that anything really changed). Unless someone wants to speak up, I'll commit this soonish (hopefully tomorrow, but at least sometime later this week). Thanks! Stephen
Attachment
Greetings, * Stephen Frost (sfrost@snowman.net) wrote: > Awesome, attached is just a rebase (not that anything really changed). > Unless someone wants to speak up, I'll commit this soonish (hopefully > tomorrow, but at least sometime later this week). Alright, as this took a bit more playing with to work cleanly in the back-branches, I'm putting it back out there with full patches for all the back-branches, in case anyone sees anything I missed, but I think I got it all right and the docs build for me and at least going through all the new pages, everything looks good to me. Naturally, only included the appropriate pieces in each of the back branches (v10 got the xlog -> WAL changes, v11 had the same, v12 had those plus the recovery.conf changes, as did v13 and HEAD). Once these all go in, I'll update the default roles patch as discussed elsewhere and backpatch that too. If there's other things we've done that would be good to include here, I'd be happy to work with anyone who's interested in putting in the effort to add more. For now, this seems like a pretty good set though. Unless there's anything further, will commit these soon. Thanks! Stephen
Attachment
On Sun, Mar 28, 2021, at 2:22 PM, Stephen Frost wrote:
Unless there's anything further, will commit these soon.
I briefly looked at this patch and have a few comments.
+<sect1 id="app-pgreceivexlog" xreflabel="pg_receivexlog">
+ <title><command>pg_receivexlog</command> renamed to <command>pg_recievewal</command></title>
s/pg_recievewal/pg_receivewal/
+<appendix id="appendix-obsolete">
+ <title>Obsolete or renamed features, settings and files</title>
Section titles are capitalized so it should be "Obsolete or Renamed Features,
Settings and Files". I find this section name too descriptive. I didn't follow
this thread but maybe we might use a generic name that is also shorter than it
such as "Incompatible or Obsolete Features".
+ <primary>trigger_file</primary>
+ <see>promote_trigger_file</see>
+ </indexterm>
+ setting has been renamed to
+ <xref linkend="guc-promote-trigger-file"/>
+ </para>
There should be a period after the <xref>.
Greetings, * Euler Taveira (euler@eulerto.com) wrote: > On Sun, Mar 28, 2021, at 2:22 PM, Stephen Frost wrote: > > Unless there's anything further, will commit these soon. > I briefly looked at this patch and have a few comments. > > +<sect1 id="app-pgreceivexlog" xreflabel="pg_receivexlog"> > + <title><command>pg_receivexlog</command> renamed to <command>pg_recievewal</command></title> > > s/pg_recievewal/pg_receivewal/ Hah! Good catch. Fixed. > +<appendix id="appendix-obsolete"> > + <title>Obsolete or renamed features, settings and files</title> > > Section titles are capitalized so it should be "Obsolete or Renamed Features, > Settings and Files". I find this section name too descriptive. I didn't follow > this thread but maybe we might use a generic name that is also shorter than it > such as "Incompatible or Obsolete Features". Not sure how others feel, but 'incompatible' doesn't seem quite right to me- we renamed some things between major versions but they're as compatible as they were before. Having it be shorter does make sense to me, so I updated the patch with "Obsolete or Renamed Features" instead. I also reworded the section underneath a bit to mention renaming and to include a comment about features, settings, and file names. > + <primary>trigger_file</primary> > + <see>promote_trigger_file</see> > + </indexterm> > + setting has been renamed to > + <xref linkend="guc-promote-trigger-file"/> > + </para> > > There should be a period after the <xref>. Good catch, fixed. Updated patches attached. Thanks! Stephen
Attachment
Greetings, * Stephen Frost (sfrost@snowman.net) wrote: > * Stephen Frost (sfrost@snowman.net) wrote: > > Awesome, attached is just a rebase (not that anything really changed). > > Unless someone wants to speak up, I'll commit this soonish (hopefully > > tomorrow, but at least sometime later this week). > > Alright, as this took a bit more playing with to work cleanly in the > back-branches, I'm putting it back out there with full patches for all > the back-branches, in case anyone sees anything I missed, but I think I > got it all right and the docs build for me and at least going through > all the new pages, everything looks good to me. > > Naturally, only included the appropriate pieces in each of the back > branches (v10 got the xlog -> WAL changes, v11 had the same, v12 had > those plus the recovery.conf changes, as did v13 and HEAD). > > Once these all go in, I'll update the default roles patch as discussed > elsewhere and backpatch that too. If there's other things we've done > that would be good to include here, I'd be happy to work with anyone > who's interested in putting in the effort to add more. For now, this > seems like a pretty good set though. > > Unless there's anything further, will commit these soon. And done. Thanks all! Stephen