bootstrap should fail gracefully, if there is no pkg-config available #27219

dimpase · 2019-02-04T20:15:58Z

When #27114 is merged, ./bootstrap requires pkgconf to run properly. It should switch to downloading the configure spkg if the pkgconf m4 macros are not installed. Otherwise one gets

checking whether g++ supports C++11 features with -std=gnu++11... yes
checking for gcc option to accept ISO C99... none needed
checking if gcc accepts -dumpversion option... yes
checking gcc version... 5.4.0
checking if g++ accepts -dumpversion option... yes
checking g++ version... 5.4.0
configure: === checking whether to install the libffi SPKG ===
./configure: line 9026: syntax error near unexpected token `LIBFFI,'
./configure: line 9026: `        PKG_CHECK_MODULES(LIBFFI, libffi, ,'
If you would like to try to build Sage anyway (to help porting),
export the variable 'SAGE_PORT' to something non-empty.
Makefile:39: recipe for target 'build/make/Makefile' failed
make[1]: Leaving directory '/var/lib/buildbot/slave/sage_git/build'
make[1]: *** [build/make/Makefile] Error 1
Makefile:31: recipe for target 'build-clean' failed
make: *** [build-clean] Error 2
program finished with exit code 2

In order to solve this issue, we check for presence of PKG_PROG_PKG_CONFIG at bootstrap/autoconf run. We need renaming of PKG_ -> SPKG_ in m4/sage_spkg_collect.m4, which we take from #27114 in order not to cause conflicts.

#27114 probably will need a rebase on top of this branch.

CC: @embray @vbraun @jdemeyer

Component: build

Author: Dima Pasechnik

Branch/Commit: 187de34

Reviewer: Jeroen Demeyer

Issue created by migration from https://trac.sagemath.org/ticket/27219

The text was updated successfully, but these errors were encountered:

dimpase · 2019-02-04T22:31:36Z

Branch: u/dimpase/build/bootstrap_check

dimpase · 2019-02-04T22:31:36Z

Commit: af57d5b

dimpase · 2019-02-04T22:31:36Z

comment:1

OK, Volker, how about this?

vbraun · 2019-02-06T23:34:39Z

comment:2

IMHO whether or not there is a pkg-config binary isn't the same as autoconf understanding PKG_CHECK_MODULES.

dimpase · 2019-02-07T01:32:48Z

comment:3

Replying to @vbraun:

IMHO whether or not there is a pkg-config binary isn't the same as autoconf understanding PKG_CHECK_MODULES.

PKG.m4 - macros are a part of pkg-config. (a sufficiently new one - we can add a version check in bootstrap...)

We can also add a call to PKG_PROG_PKG_CONFIG to configure.ac.

It is of course perfectly possible to have a mangled install of pkg-config, without the needed m4 macros.

dimpase · 2019-02-07T01:32:48Z

Author: Dima Pasechnik

jdemeyer · 2019-02-07T06:04:46Z

comment:4

Replying to @vbraun:

IMHO whether or not there is a pkg-config binary isn't the same as autoconf understanding PKG_CHECK_MODULES.

I agree. The right thing to check then is whether the macro PKG_CHECK_MODULES is expanded.

jdemeyer · 2019-02-07T06:32:58Z

comment:5

Also, what goes wrong when pkg-config is not installed? The bug report should say that.

jdemeyer · 2019-02-07T06:40:47Z

comment:6

m4_pattern_forbid may be the correct solution here. See https://www.gnu.org/software/autoconf/manual/autoconf.html#Forbidden-Patterns

sagetrac-git · 2019-02-07T10:07:22Z

Changed commit from af57d5b to 6e9f902

sagetrac-git · 2019-02-07T10:07:22Z

Branch pushed to git repo; I updated commit sha1. This was a forced push. New commits:

`6e9f902`	`check that pkg-config and its m4 macros are installed`

dimpase · 2019-02-07T10:09:34Z

comment:8

from the latest commit message:

    check that pkg-config and its m4 macros are installed
    
    use m4_ifndef() in configure.ac to check that PKG_PROG_PKG_CONFIG
    is provided.
    
    just in case, also check that pkg-config is present
    
    calling PKG_PROG_PKG_CONFIG in configure.ac
    
    renaming of clashing with PKG_* m4 names in m4/*
    (otherwise PKG_PROG_PKG_CONFIG cannot be mentioned at all)

took a while to figure out weird errors caused by m4 vars name clashes.

dimpase · 2019-02-07T10:11:27Z

comment:9

Not 100% sure whether we ought to call PKG_PROG_PKG_CONFIG. I do this in the current branch,
I suppose it cannot hurt.

vbraun · 2019-02-07T11:53:24Z

comment:10

What currently happens is

checking whether g++ supports C++11 features with -std=gnu++11... yes
checking for gcc option to accept ISO C99... none needed
checking if gcc accepts -dumpversion option... yes
checking gcc version... 5.4.0
checking if g++ accepts -dumpversion option... yes
checking g++ version... 5.4.0
configure: === checking whether to install the libffi SPKG ===
./configure: line 9026: syntax error near unexpected token `LIBFFI,'
./configure: line 9026: `        PKG_CHECK_MODULES(LIBFFI, libffi, ,'
If you would like to try to build Sage anyway (to help porting),
export the variable 'SAGE_PORT' to something non-empty.
Makefile:39: recipe for target 'build/make/Makefile' failed
make[1]: Leaving directory '/var/lib/buildbot/slave/sage_git/build'
make[1]: *** [build/make/Makefile] Error 1
Makefile:31: recipe for target 'build-clean' failed
make: *** [build-clean] Error 2
program finished with exit code 2

jdemeyer · 2019-02-07T12:22:45Z

comment:11

Right, so all that you need is something like m4_pattern_forbid([^PKG_]).

What's the point of checking for pkgconf in bootstrap? Generating the ./configure script should be orthogonal to whether or not pkgconf is installed.

sagetrac-git · 2019-02-07T12:27:54Z

Branch pushed to git repo; I updated commit sha1. New commits:

`371cd8e`	`use Eriks's m4 fixes`
`4d9ccb4`	`Trac #27109: Use pkg-config if possible to detect libffi`
`761c978`	`AC_CHECK_HEADERS should come before AC_SEARCH_LIBS`
`3d2eb36`	`move ffi/ffi.h first since it seems to be more common too`
`cc37228`	`Merge remote-tracking branch 'trac/u/embray/build/ticket-27109' into booboo`

sagetrac-git · 2019-02-07T12:27:54Z

Changed commit from 6e9f902 to cc37228

dimpase · 2019-02-07T12:43:48Z

comment:18

oops, sorry, I should have put in the old version of libffi/spkg-configure.m4...

jdemeyer · 2019-02-07T12:44:29Z

comment:19

Replying to @dimpase:

Generating configure needs pkg-config since #27219

You need to distinguish between generating configure and running configure. In this ticket, we are only concerned with the bootstrap script which runs autoconf to generate configure.

Now autoconf is a macro processor, so it would be surprising that it would run pkg-config. Even if it does, what happens if pkg-config is not installed? I would assume that it fails gracefully, so I still don't see the point of the extra pkg-config --version check.

jdemeyer · 2019-02-07T12:48:31Z

comment:20

By the way, if you want to use m4_ifndef, at least make sure that error message is correct:

m4_ifndef([PKG_PROG_PKG_CONFIG],
    [m4_fatal([Could not locate the pkg-config autoconf macros.
    These are usually located in /usr/share/aclocal/pkg.m4. If your
    macros are in a different location, try setting the environment
    variable ACLOCAL="aclocal -I/other/macro/dir" before running
    autoreconf/autogen.sh.])])

The script in Sage is called bootstrap.

jdemeyer · 2019-02-07T12:51:09Z

comment:21

Replying to @dimpase:

oops, sorry, I should have put in the old version of libffi/spkg-configure.m4...

In general, your commit history on this ticket is a mess, so I suggest to rebase and squash.

dimpase · 2019-02-07T12:57:58Z

comment:22

I am certainly distinguishing between generating and running configure. Generating configure only needs m4 macros, this is correct.

Would you like me to take out from bootstrap the test for pkg-config presence?

dimpase · 2019-02-07T13:12:22Z

comment:23

Replying to @jdemeyer:

Replying to @dimpase:

oops, sorry, I should have put in the old version of libffi/spkg-configure.m4...

In general, your commit history on this ticket is a mess, so I suggest to rebase and squash.

the problem is that for this to work, I need changes to m4/sage_spkg_collect.m4
akin to ones done on #27114. (I first did my own, without using #27114, only to find out that merging #27114 gets painful with them; so I decided to use a part of #27114, but haven't done it quite right...).

After I'm done here, #27114 will have to be rebased on top of this one.

sagetrac-git · 2019-02-07T13:32:15Z

Changed commit from 89a402f to 87621c7

sagetrac-git · 2019-02-07T13:32:15Z

Branch pushed to git repo; I updated commit sha1. This was a forced push. New commits:

`453033d`	`renaming PKG_* -> SPKG_* in m4`
`87621c7`	`check for pkg-config's m4 macros at bootstrap`

dimpase · 2019-02-07T13:39:31Z

comment:25

I hope I have addressed all the issues raised, including the messy branches. Sorry, it took a while to get right.

jdemeyer · 2019-02-07T14:08:04Z

comment:26

Indeed, this is more or less what I meant.

For the record, the message when the pkg-config macros are not installed:

configure.ac:219: error: Could not locate the pkg-config autoconf macros.
    These are usually located in /usr/share/aclocal/pkg.m4. If your
    macros are in a different location, try setting the environment
    variable ACLOCAL="aclocal -I/other/macro/dir" before running
    ./bootstrap
configure.ac:219: the top level
autom4te: /usr/bin/m4 failed with exit status: 1
aclocal-1.15: error: echo failed with exit status: 1
Bootstrap failed. Either install pkg-config m4 macros for autotools or run bootstrap with                                                                               
the -d option to download the auto-generated files instead.

So it fails correctly.

jdemeyer · 2019-02-07T14:17:31Z

comment:27

One more detail: it would be better to reserve a specific exit code for that case. I'll prepare a review commit.

jdemeyer · 2019-02-07T14:25:58Z

Changed branch from u/dimpase/build/bootstrap_check to u/jdemeyer/build/bootstrap_check

jdemeyer · 2019-02-07T14:28:29Z

Changed commit from 87621c7 to 187de34

jdemeyer · 2019-02-07T14:28:29Z

comment:29

I'm happy with this now. Feel free to set to positive review if you agree.

New commits:

`dcf2ab6`	`renaming PKG_* -> SPKG_* in m4`
`13c87f7`	`check for pkg-config's m4 macros at bootstrap`
`187de34`	`Use exit status 16 for missing pkg-config macros`

jdemeyer · 2019-02-07T14:28:29Z

Reviewer: Jeroen Demeyer

dimpase · 2019-02-07T14:40:23Z

comment:30

looks good to me, thanks!

embray · 2019-02-07T14:50:39Z

comment:31

I've been AFK, else I would've chimed in sooner. But this looks good to me now.

Indeed, there's no need to actually have pkg-config to build anything, just the PKG_ macros that are typically installed alongside pkg-config. I had also just suggested adding a copy of pkg.m4 to our m4/ directory but I like this even better (I see no reason a build machine, except maybe on OSX, shouldn't have pkg-config installed anyways).

vbraun · 2019-02-08T23:21:43Z

Changed branch from u/jdemeyer/build/bootstrap_check to 187de34

dimpase added this to the sage-8.7 milestone Feb 4, 2019

dimpase added c: build labels Feb 4, 2019

dimpase added the s: needs review label Feb 4, 2019

jdemeyer added s: needs work and removed s: needs review labels Feb 7, 2019

dimpase added s: needs review and removed s: needs work labels Feb 7, 2019

This comment has been minimized.

Sign in to view

jdemeyer added s: needs work and removed s: needs review labels Feb 7, 2019

This comment has been minimized.

Sign in to view

dimpase added the s: needs review label Feb 7, 2019

jdemeyer added s: needs work and removed s: needs review labels Feb 7, 2019

This comment has been minimized.

Sign in to view

dimpase added s: needs review and removed s: needs work labels Feb 7, 2019

dimpase added s: positive review and removed s: needs review labels Feb 7, 2019

vbraun removed the s: positive review label Feb 8, 2019

vbraun closed this as completed in 8a58b71 Feb 8, 2019

jdemeyer mentioned this issue Feb 12, 2019

Further libffi fixes #27114

Closed

embray mentioned this issue Jan 29, 2023

Meta-ticket: spkg-configure: Try to use as many system packages as possible #27330

Open

bootstrap should fail gracefully, if there is no pkg-config available #27219

bootstrap should fail gracefully, if there is no pkg-config available #27219

Comments

dimpase commented Feb 4, 2019

dimpase commented Feb 4, 2019

dimpase commented Feb 4, 2019

dimpase commented Feb 4, 2019

vbraun commented Feb 6, 2019

dimpase commented Feb 7, 2019

dimpase commented Feb 7, 2019

jdemeyer commented Feb 7, 2019

jdemeyer commented Feb 7, 2019

jdemeyer commented Feb 7, 2019

sagetrac-git mannequin commented Feb 7, 2019

sagetrac-git mannequin commented Feb 7, 2019

dimpase commented Feb 7, 2019

dimpase commented Feb 7, 2019

vbraun commented Feb 7, 2019

This comment has been minimized.

jdemeyer commented Feb 7, 2019

This comment has been minimized.

sagetrac-git mannequin commented Feb 7, 2019

sagetrac-git mannequin commented Feb 7, 2019

dimpase commented Feb 7, 2019

jdemeyer commented Feb 7, 2019

jdemeyer commented Feb 7, 2019

jdemeyer commented Feb 7, 2019

dimpase commented Feb 7, 2019

dimpase commented Feb 7, 2019

sagetrac-git mannequin commented Feb 7, 2019

sagetrac-git mannequin commented Feb 7, 2019

This comment has been minimized.

dimpase commented Feb 7, 2019

jdemeyer commented Feb 7, 2019

jdemeyer commented Feb 7, 2019

jdemeyer commented Feb 7, 2019

jdemeyer commented Feb 7, 2019

jdemeyer commented Feb 7, 2019

jdemeyer commented Feb 7, 2019

dimpase commented Feb 7, 2019

embray commented Feb 7, 2019

vbraun commented Feb 8, 2019