Non-conftest nodes involved in configure checks now get node info cleared after check. #4161
Conversation
…ow get node info cleared after check.
| @@ -84,10 +84,10 @@ | |||
| %(sig_re)s \[.*\] | |||
| conftest_%(sig_re)s_0_%(sig_re)s%(_obj)s: | |||
| %(_sconf_temp_conftest_0_c)s: %(sig_re)s \d+ \d+ | |||
| %(CC)s: %(sig_re)s \d+ \d+ | |||
| %(CC)s: %(sig_re)s None None | |||
There was a problem hiding this comment.
this is the compiler which is considered a non-conftest node, so if only configure checks are run, then the node info is not stored.
| int test_header = 2; | ||
| """) | ||
|
|
||
| test.not_up_to_date(read_str="Checking for C header file header1.h... (cached) yes\n") |
There was a problem hiding this comment.
I tested before and after and verified this test fails with master, and passed with the PR
|
Not sure about the sider issue, should the code style remain consistent? |
|
What happens when you run scons twice in a row where a real file is used by configure, but nothing depends on it? |
punt on this one. There'd be too much to fix.. We'll do a black run on the whole codebase at some point and then enforce. |
No it uses the cached value. I updated the test to show this. |
mwichmann
changed the title
|
@dmoody256 - so from looking at the code, I think the only possible functional change is that if you were to only run the configure step and no actual build steps the sconsign would be missing the content signature information for all non-configure generated files uses while processing the configure checks? Does that sound correct? |
Yes |
Fixes #2757
from https://pairlist4.pair.net/pipermail/scons-users/2022-May/008989.html
Contributor Checklist:
CHANGES.txt(and read theREADME.rst)