Thread: [HACKERS] building libpq.a static library
I maintain static libraries for libpq for the R programming language (we need static linking to ship with the binary packages). Unfortunately currently the standard postgres makefile only generates a shared library for libpq, not a static one. In order to make a static library I always manually edit ./src/interfaces/libpq/Makefile and add a target: libpq.a: $(OBJS) ar rcs $@ $^ And then run 'make libpq.a' in the libpq directory. This works but it's a bit of a pain to maintain. I was wondering if this hack could be merged so that the standard 'configure --enable-static' script would install a static library for libpq alongside the shared one.
On Wednesday, July 12, 2017 6:31:09 AM EDT Jeroen Ooms wrote: > I maintain static libraries for libpq for the R programming language > (we need static linking to ship with the binary packages). > Unfortunately currently the standard postgres makefile only generates > a shared library for libpq, not a static one. > > In order to make a static library I always manually edit > ./src/interfaces/libpq/Makefile and add a target: > > libpq.a: $(OBJS) > ar rcs $@ $^ > > And then run 'make libpq.a' in the libpq directory. > > This works but it's a bit of a pain to maintain. I was wondering if > this hack could be merged so that the standard 'configure > --enable-static' script would install a static library for libpq > alongside the shared one. I have no idea what the consensus about this is, but providing a patch file and registering it at https://commitfest.postgresql.org/ increases your chances. I wish you good luck hacking autoconf hell (which is what you'll need to do to get the command line switch) though (*shudder*).
Jeroen Ooms <jeroen@berkeley.edu> writes: > I maintain static libraries for libpq for the R programming language > (we need static linking to ship with the binary packages). How do you get that past vendor packaging policies? When I worked at Red Hat, there was a very strong policy against allowing any package to statically embed parts of another one, because it creates serious management problems if e.g. the other one needs a security update. I'm sure Red Hat isn't the only distro that feels that way. I think you'd be better advised to fix things so you can link with the standard shared-library version of libpq (and whatever else you're doing this with). > This works but it's a bit of a pain to maintain. I was wondering if > this hack could be merged so that the standard 'configure > --enable-static' script would install a static library for libpq > alongside the shared one. FWIW, we used to have support for building static libpq, but we got rid of it a long time ago. I couldn't find the exact spot in some desultory trawling of the commit history. regards, tom lane
On Wed, Jul 12, 2017 at 5:11 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote: > Jeroen Ooms <jeroen@berkeley.edu> writes: >> I maintain static libraries for libpq for the R programming language >> (we need static linking to ship with the binary packages). > > How do you get that past vendor packaging policies? When I worked at > Red Hat, there was a very strong policy against allowing any package > to statically embed parts of another one, because it creates serious > management problems if e.g. the other one needs a security update. > I'm sure Red Hat isn't the only distro that feels that way. We only use this on Windows. On platforms with a decent package manager we indeed link to a shared library. > FWIW, we used to have support for building static libpq, but > we got rid of it a long time ago. OK that's too bad. I agree that shared libs are often preferable but in some environments dynamic linking is simply not possible and you need to static link the library into the application. Most C/C++ libraries do support --enable-static and even for most linux distros the static libs are included with the lib-dev / lib-devel package.
On 12 July 2017 at 16:11, Tom Lane <tgl@sss.pgh.pa.us> wrote: > Jeroen Ooms <jeroen@berkeley.edu> writes: > >> This works but it's a bit of a pain to maintain. I was wondering if >> this hack could be merged so that the standard 'configure >> --enable-static' script would install a static library for libpq >> alongside the shared one. > > FWIW, we used to have support for building static libpq, but > we got rid of it a long time ago. I couldn't find the exact > spot in some desultory trawling of the commit history. Fwiw I think the real problem is that building static libraries "properly" requires different compiler options -- notably they're not normally built with -fPIC. So that means building every object twice which kind of breaks make's build model which has a simple dependency graph where each object appears once. Some packages do this by inventing a foo-shared.o and foo-static.o but that introduces its own weirdness. I don't know what the downsides would be of creating a static library out of objects built with -fPIC. It might just be a small performance penalty which might be no big deal for libpq. That may be a good compromise. -- greg
On 2017-07-12 23:55:56 +0100, Greg Stark wrote: > Fwiw I think the real problem is that building static libraries > "properly" requires different compiler options -- notably they're not > normally built with -fPIC. So that means building every object twice > which kind of breaks make's build model which has a simple dependency > graph where each object appears once. Some packages do this by > inventing a foo-shared.o and foo-static.o but that introduces its own > weirdness. > > I don't know what the downsides would be of creating a static library > out of objects built with -fPIC. It might just be a small performance > penalty which might be no big deal for libpq. That may be a good > compromise. FWIW, most linux distributions build everything with -fPIC/PIE anyway these days, to allow address space randomization. So I don't think this is a huge concern for modern platforms. - Andres
On 12 July 2017 at 23:46, Jeroen Ooms <jeroen@berkeley.edu> wrote:
On Wed, Jul 12, 2017 at 5:11 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Jeroen Ooms <jeroen@berkeley.edu> writes:
>> I maintain static libraries for libpq for the R programming language
>> (we need static linking to ship with the binary packages).
>
> How do you get that past vendor packaging policies? When I worked at
> Red Hat, there was a very strong policy against allowing any package
> to statically embed parts of another one, because it creates serious
> management problems if e.g. the other one needs a security update.
> I'm sure Red Hat isn't the only distro that feels that way.
We only use this on Windows. On platforms with a decent package
manager we indeed link to a shared library.
You shouldn't ever need static libraries on Windows, though. Because it searches the CWD first on its linker search path, you can just drop libpq.dll in the same directory as your binary/library and link to the stub libpq.lib .
On Mac OS X and Linux, you can use relative rpath linking. On OS X use @executable_path either at link-time or later via install_name_tool . On Linux, use $ORIGIN in your rpath. Beware of quoting issues with $ORIGIN though.
I'm not trying to block support for a static libpq, I just don't see the point.
On 13 July 2017 at 10:58, Craig Ringer <craig@2ndquadrant.com> wrote:
-- On 12 July 2017 at 23:46, Jeroen Ooms <jeroen@berkeley.edu> wrote:On Wed, Jul 12, 2017 at 5:11 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Jeroen Ooms <jeroen@berkeley.edu> writes:
>> I maintain static libraries for libpq for the R programming language
>> (we need static linking to ship with the binary packages).
>
> How do you get that past vendor packaging policies? When I worked at
> Red Hat, there was a very strong policy against allowing any package
> to statically embed parts of another one, because it creates serious
> management problems if e.g. the other one needs a security update.
> I'm sure Red Hat isn't the only distro that feels that way.
We only use this on Windows. On platforms with a decent package
manager we indeed link to a shared library.You shouldn't ever need static libraries on Windows, though. Because it searches the CWD first on its linker search path
Er, sorry, binary location, not CWD.
On Thu, Jul 13, 2017 at 4:58 AM, Craig Ringer <craig@2ndquadrant.com> wrote: > You shouldn't ever need static libraries on Windows, though. Because it > searches the CWD first on its linker search path, you can just drop > libpq.dll in the same directory as your binary/library and link to the stub > libpq.lib . This is not possible in our case. The R programming language binaries are installed in the "program files" directory which is only writable by the sys admin. There are over 10.000 contributed packages (including one with postgres drivers) which are installed by the user in the home dir and the package DLL's need to get dynamically loaded at runtime. We have been working with this system for a very long time and static linking external libs to the package DLL is really the only reliable way to prevent DLL problems across Windows versions/configurations > I'm not trying to block support for a static libpq, I just don't see the > point. This is a bit overgeneralized, there are many use cases where static linking is the only feasible way. Most libs support --enable static and many distros ship both static and shared libs and leave it up to user or developer author how to configure their application.
On 7/12/17 11:11, Tom Lane wrote: > FWIW, we used to have support for building static libpq, but > we got rid of it a long time ago. I couldn't find the exact > spot in some desultory trawling of the commit history. We still build and install static libraries. -- Peter Eisentraut http://www.2ndQuadrant.com/ PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services
Peter Eisentraut <peter.eisentraut@2ndquadrant.com> writes: > On 7/12/17 11:11, Tom Lane wrote: >> FWIW, we used to have support for building static libpq, but >> we got rid of it a long time ago. I couldn't find the exact >> spot in some desultory trawling of the commit history. > We still build and install static libraries. Hm, I'd taken Jeroen's complaint at face value, but that sure explains why I couldn't find where it'd been removed ;-) The answer may then be that he's working with a vendor-packaged version and the vendor chose to enforce their distro policy about not shipping static libraries by patching that build step out of libpq's Makefile. regards, tom lane