configure shouldn't set a default OPT

[pitrou]

BPO	10663
Nosy	@warsaw, @pitrou, @merwok, @davidmalcolm, @iritkatriel

^{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 = None
created_at = <Date 2010-12-09.19:25:28.138>
labels = ['type-bug', '3.10', 'build']
title = "configure shouldn't set a default OPT"
updated_at = <Date 2021-04-17.04:43:11.735>
user = 'https://github.com/pitrou'

bugs.python.org fields:

activity = <Date 2021-04-17.04:43:11.735>
actor = 'Arfrever'
assignee = 'none'
closed = False
closed_date = None
closer = None
components = ['Build']
creation = <Date 2010-12-09.19:25:28.138>
creator = 'pitrou'
dependencies = []
files = []
hgrepos = []
issue_num = 10663
keywords = []
message_count = 4.0
messages = ['123695', '135468', '379807', '391271']
nosy_count = 7.0
nosy_names = ['barry', 'pitrou', 'eric.araujo', 'Arfrever', 'laca', 'dmalcolm', 'iritkatriel']
pr_nums = []
priority = 'normal'
resolution = None
stage = None
status = 'open'
superseder = None
type = 'behavior'
url = 'https://bugs.python.org/issue10663'
versions = ['Python 3.10']

Linked PRs

• gh-54872: Remove default OPT=-O for non-GCC compilers #142159

[pitrou]

The configure.in sets a default OPT of "-O" if none was set by the user, but I think that's wrong. The user could simply pass optimization flags as part of CFLAGS instead, and then the contents of OPT could conflict with that of CFLAGS (which is annoying to debug when you don't know what is happening exactly).

Besides, "-O" is hardly an useful default value for any compiler. I would advocate trimming down the magic and letting OPT empty/undefined if that's what the user asks for (and expects). What do you think?

[merwok]

Sounds reasonable.

[iritkatriel]

I couldn't find a configure.in. Is this issue out of date?

[Arfrever]

configure.in is old name of Autoconf input file.
configure.ac is new name of Autoconf input file (and the only name to be supported in some future version of Autoconf).
configure.in in CPython project was renamed to configure.ac in 2012 (commit 0f4c16e).
Problem described in this issue is still present in configure.ac.