make libs/readline know where headers/libs are

[dimpase]

To allow non-standard locations for readline (e.g. with MacOS Homebrew)
one should use # distutils: include_dirs = etc in sage/libs/readline.pxd to make it look in non-standard locations.

readline/spkg-configure.m4 should set up appropriate variables
to be filled in and exported in src/bin/sage-env-config.in; in turn
they should make it into cython_aliases in sage/env.py.

This came up in this thread on sage-devel.

See also (readline-related):

• qepcad build fails on slackware-14.2-maximal #29573 qepcad build fails on slackware-14.2-maximal
• Fix perl_term_readline_gnu build error on conda-forge-macos-maximal #29572 Fix perl_term_readline_gnu build error on conda-forge-macos-maximal

See also (similar failure mode):

• homebrew-standard: Fix several problems with library search paths #29607 comment 45
• configure finds libpng but matplotlib does not #29448

Depends on #32073

CC: @dimpase @dcoudert @isuruf @orlitzky

Component: build: configure

Branch/Commit: public/packages/readline_underlink @ af7cec6

Reviewer: Dima Pasechnik

Issue created by migration from https://trac.sagemath.org/ticket/29000

[mwageringel]

comment:1

Also note that several packages seem to link against the macOS readline instead of the Homebrew one. In fact, the only one I am sure links against the Homebrew readline is rpy2. What would be needed to solve this? The Homebrew version provides a pkg-config file, in case that is useful.

I am hesitant to permanently install Homebrew's readline to /usr/local as this is against what is intended by Homebrew. In the case of readline though, I am not sure how much of a problem this really is.

[dimpase]

comment:2

GAP needs a "real" readline, not a MacOS's libedit.

IIRC, sagelib's readline also needs "real" readline.

I did try installing Homebrew's readline into /usr/local, it seems to cause no harm, and
allowed builds to work on MacOS 10.13.

[mwageringel]

comment:3

Replying to @dimpase:

GAP needs a "real" readline, not a MacOS's libedit.

This might explain why I currently see several persistent doctest timeouts with this setup. Giac seems to be affected, as well. I am a bit surprised the build does not fail with this incompatible readline (other than for sage.libs.readline).

[mwageringel]

Description changed:

--- 
+++ 
@@ -4,3 +4,5 @@
 `readline/spkg-configure.m4` should set up appropriate variables
 to be filled in and exported in `src/bin/sage-env-config.in`; in turn
 they should make it into `cython_aliases` in `sage/env.py`. 
+
+This came up in this [thread on sage-devel](https://groups.google.com/d/msg/sage-devel/91e0GfHT-Yo/Vvj8X8ESCgAJ).

[mkoeppe]

comment:5

Thanks for adding the link to the problem description.

What is happening here is that you are relying on an automake feature that our build system (unfortunately) does not have.

It is true that one can set variables by passing them to configure, and they will be respected during the configure run and remembered in config.status. However, in our build system they are not saved in any configure-generated output file:

$ ./configure LDFLAGS="-L/xyzzy1" CPPFLAGS="-I/xyzzy2" PKG_CONFIG_PATH="/xyzzy3"
$ grep xyzzy build/make/Makefile-auto build/make/Makefile src/Makefile src/bin/sage-env-config build/bin/sage-build-env-config build/pkgs/sage_conf/src/sage_conf.py build/pkgs/sage_conf/src/setup.cfg

... except in the automake-generated file build/make/Makefile-auto, which is not used at all in our build process.

build/make/Makefile-auto:CPPFLAGS =  -I/xyzzy2
build/make/Makefile-auto:LDFLAGS = -L/xyzzy1
build/make/Makefile-auto:PKG_CONFIG_PATH = /xyzzy3

[mwageringel]

comment:6

Thank you for explaining this. This clears up a lot. I think I will avoid the keg-only packages for now.

[mkoeppe]

comment:7

Also PARI's spkg-install.in hardcodes readline to come from SAGE_LOCAL

bash ./Configure --prefix="$SAGE_LOCAL" \
    --with-readline="$SAGE_LOCAL" $SAGE_CONFIGURE_GMP \
    --with-runtime-perl="/usr/bin/env perl" \
    --kernel=gmp $PARI_CONFIGURE 

which is not correct when readline/spkg-configure.m4 found a system readline.

On local-homebrew-macos-standard (#29417), this leads to:

Hi-Res Graphics: none
###
### Editline wrapper detected, building without readline support
###
### Building without GNU readline support

[dimpase]

comment:8

I believe these are missing for packages for which readline is not a dependency (but a feature). On the other hand, perhaps it can potentially pollute the compiling/linking calls.

Let's fix it...

[mkoeppe]

comment:10

Another case:

• Fix perl_term_readline_gnu build error on conda-forge-macos-maximal #29572 Fix perl_term_readline_gnu build error on conda-forge-macos-maximal

[dimpase]

Author: Dima Pasechnik

[mkoeppe]

Description changed:

--- 
+++ 
@@ -6,3 +6,7 @@
 they should make it into `cython_aliases` in `sage/env.py`. 
 
 This came up in this [thread on sage-devel](https://groups.google.com/d/msg/sage-devel/91e0GfHT-Yo/Vvj8X8ESCgAJ).
+
+
+See also: 
+- #29573 qepcad build fails on slackware-14.2-maximal