Re: bmake inefficiencies

["Simon J. Gerraty" <sjg%juniper.net@localhost>]

Mateusz Guzik <mjguzik%gmail.com@localhost> wrote:

> [External Email. Be cautious of content]
> 
> 
> Responding to the original mail due to several follow ups.
> 
> Thanks for fixups so far.
> 
> Another note related to JobOpenTmpFile:
> 
>         JobSigLock(&mask);
>         tfd = mkTempFile(TMPPAT, &tfile);
>         if (!DEBUG(SCRIPT))
>                 (void)eunlink(tfile);
>         JobSigUnlock(&mask);
> 
> is the signal stuff really necessary?

It avoids the need to loop dealing with interupts.
The name seems missleading should be block not lock.

> Apart from that, eunlink stats the path and this has happens in
> default setting. Instead I propose the following: since mkTempFile has
> only 2 consumers, passing in the buffer can be made mandatory. Then
> another flag can control when the routine should unlink the result.
> This would shorten this callsite to roughly:
> 
> char buf[MAXPATHLATEN];
> ...
> tfd = mkTempFile(TMPPAT, buf, sizeof buf, !DEBUG(SCRIPT));
> 
> which will also save on memory alloc and string coying as the target
> routine calls bmake_strdup to accomodate the request.

Will take a look.