gh-90005: Port readline and curses to PY_STDLIB_MOD (GH-94452)

[tiran]

• Issue: Port module setup to PY_STDLIB_MOD() macro and addext() #90005

[bedevere-bot]

🤖 New build scheduled with the buildbot fleet by @tiran for commit 33d9e296124d7daeef89fca957bdab51e1e1d11b 🤖

If you want to schedule another build, you need to add the ":hammer: test-with-buildbots" label again.

[bedevere-bot]

🤖 New build scheduled with the buildbot fleet by @tiran for commit dd5937322e181781c03213621eddb0fb88527c48 🤖

If you want to schedule another build, you need to add the ":hammer: test-with-buildbots" label again.

[erlend-aasland]

Great work! I left some comments.

[erlend-aasland]

Looks like this will create havoc for WASM/Emscripten static builds; multiple -l switches are passed to the compiler:

config.log:

29899 configure:21034: checking for rl_pre_input_hook in -lreadline
29900 configure:21059: gcc -o conftest     conftest.c -lreadline  -lreadline -lintl -ldl  >&5
29901 configure:21059: $? = 0
29902 configure:21069: result: yes
29903 configure:21080: checking for rl_completion_display_matches_hook in -lreadline
29904 configure:21105: gcc -o conftest     conftest.c -lreadline  -lreadline -lintl -ldl  >&5
29905 configure:21105: $? = 0
29906 configure:21115: result: yes
29907 configure:21126: checking for rl_resize_terminal in -lreadline
29908 configure:21151: gcc -o conftest     conftest.c -lreadline  -lreadline -lintl -ldl  >&5

[tiran]

Good point, should I place a WITH_SAVE_ENV around each check? By the way, readline is not yet supported on Emscripten.

[erlend-aasland]

WITH_SAVE_ENV can't be nested :( But your LIBS_SAVE workaround should work well.

[erlend-aasland]

Unfortunately, it does not work. They're still there.

[erlend-aasland]

AC_CHECK_LIB might be adding the library before running the check.

[tiran]

Yeah, we need a different approach here.

[erlend-aasland]

Yeah. This reminds me I've got an open PR for the exact same issue for sqlite3 dependency detection :) I've tried various approaches, but I haven't found a neat way to solve it yet. (I haven't looked at it for a week either, though.)

[erlend-aasland]

These suggestions should work. You should be able to apply them as a batch.

[tiran]

@erlend-aasland I have refactored the code. There is a bit of repetition but IMHO it's ok for a few lookups. We could add our own variant of AC_CHECK_LIB at a later point in time.

[tiran]

I have merged the PR. It works on all tested platforms. We can review and address the general issue with AC_CHECK_LIB in a future PR.

[vstinner]

The change triggered a reference leak on Fedora: #94644

[kulikjak]

Hi, I think I found an issue.

On Solaris, even a simple code like:

char readline ();
int main (void) {
  return readline ();
}

needs to be linked with both -lreadline and -lncurses; that means that checking for readline in -lreadline fails. I thought that passing LIBREADLINE_LIBS would fix that, but the content of that variable is not used until the test passes, and LIBS="-lreadline $LIBS" is always used. (AFAICT, automatic pkg-config would also do nothing, although I didn't test that).

Is there something I am missing or something I can try?

[tiran]

The problem should be fixed by GH-94802.

[kulikjak]

Ah, it's indeed fixed. I am sorry, I thought I am looking into latest main but it was 12 days old (when this was pushed).

Thanks for help!