It will be helpful for diff-ing to restrict the length of lines in the
SGML files to 71 characters (as usual).
I did it that way for the following reasons
1. It aids grep-ability
2. The committers seem to be moving towards that for SQL strings, mostly for reason #1
3. I recall that the code is put through a linter as one of the final steps before release, I assumed that the SGML gets the same.
4. Even if #3 is false, its easy enough to do manually for me to do for this one file once we've settled on the text of the definitions.
As for the changes, most things seem fine, I specifically like:
* Checkpoint - looks good
* yes, PGDATA should have been a literal
* Partition - the a/b split works for me
* Unlogged - it reads better
I'm not so sure on / responses to your ???s:
* The statement that names of schema objects are unique isn't strictly true, just mostly true. Take the case of a unique constraints. The constraint has a name and the unique index has the same name, to the point where adding a unique constraint using an existing index renames that index to conform to the constraint name.
* Serializable "other way around" question - It's both. Outside the transaction you can't see changes made inside another transaction (though you can be blocked by them), and inside serializable you can't see any changes made since you started. Does that make sense? Were you asking a different question?
* Transaction - yes, all those things could be "visible" or they could be "side effects". It may be best to leave the over-simplified definition in place, and add a "For more information see <<linref to tutorial-transactions>>
