[Bug pch/14137] [pch] ICE in cgraph_finalize_compilation_unit, at cgraphunit.c:407

["geoffk at desire dot geoffk dot org" <gcc-bugzilla@gcc.gnu.org>]

------- Additional Comments From geoffk at desire dot geoffk dot org  2004-03-27 06:16 -------
Subject: Re:  [pch] ICE in cgraph_finalize_compilation_unit, at cgraphunit.c:407


> While not all flags affect code generation

That's not the test for whether a flag will cause problems with PCH.
A good test is "If this flag could be controlled by a #pragma at the
top level (that is, outside declarations), would it work?"

> Seriously though, how likely is it that people will use different flags
> for their PCH than for their normal compilation?

What about for their abnormal compilation, or if the PCH file is not
"their" PCH?  One kind of case I'm thinking of is when someone has
just one file that they need to compile with -fno-some-optimisation,
since that optimisation turns out to make this file run slower (or
crash the compiler or whatever).  Or they have a flag like
-fpermissive that they used to use everywhere, and are trying to
eliminate, and they're about half-done.

> Doesn't Joe User typicaly just use "-O[123] -g"?

No, users as a group don't typically use any particular flag or
combination.

> I'd rather be safe and make sure all flags match than chasing flag bugs
> for PCH...

That would make PCH less useful.  You could just remove PCH
altogether, then you would have no PCH bugs to chase at all.

What I think would be helpful would be to update the documentation
every time a flag is discovered to be safe or unsafe for PCH.  At the
moment, it basically says "I don't know".  I think I will create the
structure for this now.



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14137