SConf marks files up-to-date while using them in checks

[bdbaddog]

This issue was originally created at: 2011-04-23 13:50:13.
This issue was reported by: garyo.
garyo said at 2011-04-23 13:50:13

If you have a Configure (aka SConf) test that uses a header file, say foo.h, and foo.h is modified, what happens is the Configure test runs (it's like a mini-SCons run) and updates the ninfo.csig of foo.h's node. Then the rest of the SConstruct gets read, and foo.h appears unmodified, so none of its dependents get built. Don't have a repro case handy here, will try to provide one soon.

[mwichmann]

#2485 is an example of this, it looks like.

[mwichmann]

See also this email thread: http://four.pairlist.net/pipermail/scons-users/2021-April/008527.html

[dmoody256]

Should SConf use a separate sconsign file and just stick in sconf_temp dir?

Not sure if that will really fix it, because the state may be in memory from just running the SConf checks.

[mwichmann]

That, on the surface, sounds like it could alleviate several of the interaction problems that people seem to have run into - there are a LOT of Configure bugs (of course the devil is in the details as you note). Seems like a lot of the old bugs were handwaved away with "when we rewrite blah blah...." which didn't happen.

Funny thing is, I was just thinking about another reason why one might want to be able to specify an alternate sconsign file.