This is the mail archive of the mailing list for the GCC project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [patch] Remove process IDs from temprorary file names (Nathanael Nerode) writes:

> This has been causing all kinds of trouble with my attempted top level
> bootstrap.  Is this OK for mainline?  Although not technically a bugfix,
> it avoids some Future Evil.
> What do people think about this?  And why was it being done anyway?....

Zack Weinberg wrote:

>IIRC there was a problem involving two different threads of a parallel
>build trying to execute this rule at the same time.  I do not know why
>this would happen.  However, your patch would cause it to break again.
>(Maybe the scenario was simultaneous builds for different targets with
>the same source directory?)
>Given that these files are now generated into the build directory, the
>entire problem may have been wiped out of existence.  I'd ask that you
>talk to Kelley Cook (who got rid of the generated files in the source
>directory) and Kaveh (who introduced this construct in the first place).

Accordingly, I'm talking to you.  :-)  Is it safe to eliminate the process
IDs so that the temporary files always have the same names, so that the
debug info always looks the same?

The patch in question is at

I do know of a way to avoid this causing trouble for toplevel bootstrap,
but it requires establishing a $(parsedir) separate from the build dir
and from $(srcdir) for every build, which is a pain and error-prone.

Incidentally, Kelley, does your revised scheme for these files make a
difference to this?...

(As a side question, why are the temporary files necessary?)

Nathanael Nerode  <neroden at>

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]