gh-89640: properly detect float word ordering on Linux

[erlend-aasland]

Prerequisite: python/cpython-devcontainers#23

autoconf-archive patch by Dan Amelang.

• Issue: configure script cannot detect float word ordering on linux #89640

[erlend-aasland]

@damelang, FYI I get this error on WASI:

grep: conftest.wasm: No such file or directory
grep: conftest.wasm: No such file or directory

At first, I thought using $EXEEXT was incorrect, and swapped it with .$ac_objext in 2465535, but my guess was wrong. Since the grep block is run only in case of success, I don't really see any other failure than the macro trying to grep the incorrect file name. Thoughts? I'll try to set up a WASM environment and dig deeper.

[damelang]

@damelang, FYI I get this error on WASI:

grep: conftest.wasm: No such file or directory
grep: conftest.wasm: No such file or directory

At first, I thought using $EXEEXT was incorrect, and swapped it with .$ac_objext in 2465535, but my guess was wrong. Since the grep block is run only in case of success, I don't really see any other failure than the macro trying to grep the incorrect file name. Thoughts? I'll try to set up a WASM environment and dig deeper.

Hi Erlend, I may be stating the obvious, but it looks like your use of autoconf with a WASM compiler (clang?) is misconfigured. $EXEEXT is set to ".wasm" somewhere in the configuration, but the compiler is generating a binary with something else as the extension (maybe ".wasi", or quite possible no extension at all, in which case $EXEEXT maybe should be set to the empty string?).

I've never built an autoconf-based project with a WASM compiler before, and maybe it's not common enough of a situation for others to have run into this issue. It's not a great time for me to dive into this, but if I did, I'd poke around to see where $EXEEXT is set to ".wasm". And I'd look at what extension the compiler is actually giving the WASI binaries. One of those hopefully makes more sense to use than the other, and you can simply update the configuration/code to only use that one.

I wish I knew internals well enough to tell you where to go poking around 🤷🏼‍♂️

[erlend-aasland]

It's not a great time for me to dive into this, but if I did, I'd poke around to see where $EXEEXT is set to ".wasm".

I've got a semi-functional WASI dev environment up and running with GitHub Codespaces, and I notice a few things:

• in configure.ac, we mess up EXEEXT around L1323-1340; this code ends up around L7300 in the generated configure script, several thousands of lines after the compiler and linker helpers are defined 🤦
• using --with-suffix=no works ✅
• mirroring EXEEXT to ac_exeext after our configure.ac EXEEXT hacks in L1323-1340 works ✅

Seems our EXEEXT hacks are the source of the problem.

[erlend-aasland]

Seems our EXEEXT hacks are the source of the problem.

From the top of my head, I see two options:

• in configure.ac after L1340, set ac_exeext=$EXEEXT (adding yet another hack)
• don't use EXEEXT; replace it with something outside of Autoconf's namespace, for example EXE_SUFFIX

Let's go for the latter. I'll create an issue and a PR.

[miss-islington-app]

Sorry, @erlend-aasland, I could not cleanly backport this to 3.13 due to a conflict.
Please backport using cherry_picker on command line.

cherry_picker 26d627779f79d8d5650fe7be348432eccc28f8f9 3.13

[miss-islington-app]

Sorry, @erlend-aasland, I could not cleanly backport this to 3.12 due to a conflict.
Please backport using cherry_picker on command line.

cherry_picker 26d627779f79d8d5650fe7be348432eccc28f8f9 3.12

[erlend-aasland]

If we are to backport this, we need to first backport #124657.

[hoodmane]

Hmm this change broke float word order detection on Emscripten, because EEXT is conftest.js which does not contain the floats but the file with the float order is conftest.wasm. Presumably we can hard code float word order is little endian for Emscripten, though it would be nice if autoconf knew this...

[hoodmane]

What about checking both the object file and the executable and setting the endian order if a match is found in either?

[erlend-aasland]

@hoodmane, is there an Emscripten buildbot? How can I reproduce the failure?

[hoodmane]

No, I'm working on setting up a buildbot for Emscripten again after the steering council approved switching adding back tier 3 support for it last week.

Applying this patch to aclocal.m4 and running autoconf fixes the problem:

--- a/aclocal.m4
+++ b/aclocal.m4
@@ -112,10 +112,10 @@ int main (int argc, char *argv[])
 
 ]])], [
 
-if grep noonsees conftest$EXEEXT >/dev/null ; then
+if grep noonsees conftest$EXEEXT >/dev/null || grep noonsees conftest.$ac_objext >/dev/null; then
   ax_cv_c_float_words_bigendian=yes
 fi
-if grep seesnoon conftest$EXEEXT >/dev/null ; then
+if grep seesnoon conftest$EXEEXT >/dev/null || grep seesnoon conftest.$ac_objext >/dev/null; then
   if test "$ax_cv_c_float_words_bigendian" = unknown; then
     ax_cv_c_float_words_bigendian=no
   else

For now, to reproduce:

1. Check out this branch https://github.com/hoodmane/cpython/tree/emscripten-autoconf-float-word-order
2. Get emcc on the path with:

   git clone https://github.com/emscripten-core/emsdk --depth 1
   ./emsdk/emsdk install 3.1.68
   ./emsdk/emsdk activate 3.1.68
   source ./emsdk/emsdk_env.sh

3. Use the script I added on my branch to build the "build python":

   python3.12 ./Tools/wasm/emscripten.py configure-build-python
   python3.12 ./Tools/wasm/emscripten.py make-build-python

4. Use the script to configure the "host python". This should fail with the error I indicated.

   python3.12 ./Tools/wasm/emscripten.py configure-host

[hoodmane]

Nevermind, the patch doesn't work because AC_LINK_IFELSE compiles and links in one command so there is no object file in the file system to grep. I'm not sure why I thought it worked.

[erlend-aasland]

@hoodmane, I can confirm that this breaks Emscripten. I'll reopen the issue. Perhaps a better solution is to directly manipulate BUILDEXEEXT instead of EXEEXT.

[hoodmane]

What does BUILDEXEEXT mean? There's a comment about on case insensitive platforms the suffix can't be empty? Not clear to me how it helps here, but we could add

AS_CASE([$ac_sys_system],
   [Emscripten], [EXTTOINSPECT=.wasm],
   [EXTTOINSPECT=$EXEEXT]
)

and then look at conftest$EXTTOINSPECT. Maybe I'll open a PR doing this?

[erlend-aasland]

What does BUILDEXEEXT mean?

It's the suffix the Python executable gets in the Makefile.