--with-pydebug has no effect when the final python binary is compiled

[aleb]

BPO	24116
Nosy	@smontanaro, @bitdancer, @methane, @csabella

^{Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.}

Show more details

GitHub fields:

assignee = None
closed_at = <Date 2021-12-13.18:49:20.427>
created_at = <Date 2015-05-03.07:14:38.696>
labels = ['build']
title = '--with-pydebug has no effect when the final python binary is compiled'
updated_at = <Date 2021-12-13.18:49:20.426>
user = 'https://bugs.python.org/aleb'

bugs.python.org fields:

activity = <Date 2021-12-13.18:49:20.426>
actor = 'iritkatriel'
assignee = 'none'
closed = True
closed_date = <Date 2021-12-13.18:49:20.427>
closer = 'iritkatriel'
components = ['Build']
creation = <Date 2015-05-03.07:14:38.696>
creator = 'aleb'
dependencies = []
files = []
hgrepos = []
issue_num = 24116
keywords = []
message_count = 7.0
messages = ['242455', '242456', '242458', '242468', '242553', '339891', '339926']
nosy_count = 6.0
nosy_names = ['skip.montanaro', 'Arfrever', 'r.david.murray', 'methane', 'aleb', 'cheryl.sabella']
pr_nums = []
priority = 'normal'
resolution = 'wont fix'
stage = 'resolved'
status = 'closed'
superseder = None
type = 'compile error'
url = 'https://bugs.python.org/issue24116'
versions = ['Python 3.4']

[aleb]

This is how I configure and build:

$ ./configure --prefix=/usr \
              --with-threads \
              --with-computed-gotos \
              --enable-ipv6 \
              --with-system-expat \
              --with-dbmliborder=gdbm:ndbm \
              --with-system-ffi \
              --with-system-libmpdec \
              --enable-loadable-sqlite-extensions \
              --without-ensurepip \
              --with-pydebug

$ make "CFLAGS=-g -fno-inline -fno-strict-aliasing -O0"
...
gcc -pthread -c -Wno-unused-result -g -O0 -Wall -Wstrict-prototypes -march=x86-64 -mtune=generic -O2 -pipe -fstack-protector-strong --param=ssp-buffer-size=4 -g -fno-inline -fno-strict-aliasing -O0  -Werror=declaration-after-statement   -I. -IInclude -I./Include -D_FORTIFY_SOURCE=2 -D_FORTIFY_SOURCE=2  -DPy_BUILD_CORE -o Modules/python.o ./Modules/python.c
...
gcc -pthread -Wl,-O1,--sort-common,--as-needed,-z,relro -Wl,-O1,--sort-common,--as-needed,-z,relro -Xlinker -export-dynamic -o python Modules/python.o libpython3.4dm.a -lpthread -ldl  -lutil   -lm  

Notice the specified CFLAGS are used when building "-o Modules/python.o" for example (as many others), but are not used when building the "-o python" binary.

[Arfrever]

Set CFLAGS when calling configure, not make.

[aleb]

It's confusing that the CFLAGS specified when running make are passed further sometimes but not always. So I guess that's a workaround? Or CFLAGS should never be specified to make directly?

[bitdancer]

Sounds like this is a duplicate of bpo-9189? (--with-pydebug certainly works for me).

[smontanaro]

It's confusing that the CFLAGS specified when running make are passed further sometimes but not always. So I guess that's a workaround? Or CFLAGS should never be specified to make directly?

In my experience, all the magic happens in configure. Most of the time, the only flag I give to make is -j to specify parallelism.

Is there somewhere in the build docs this could be made clearer?

[csabella]

If this is a documentation issue on the build, then I think this issue should be moved to the devguide repo on GitHub. I'm not sure if it's still unclear though, so maybe this could just be closed?

[methane]

gcc is not compiler, but linker when -o python.
So LDFLAGS is used instead of CFLAGS.

I don't think we can document and maintain all configure & make behavior.
I'm +1 on close this.