Re: 10-fyi-m4-backend.patch

[Paul Eggert]

> From: Akim Demaille <address@hidden>
> Date: 05 Feb 2002 13:10:55 +0100
> 
> Paul, is tried to avoid popen/pclose for performance issues, when
> running m4.  Would you say, that's a pointless optimization and just
> use popen,

No, it's not pointless.  It's more than just a performance issue:
popen has its own set of porting problems (e.g. with respect to signal
handling), and they are best avoided if possible.

> or, should readpipe fallback to popen/pclose when
> pipe/fork are not supported?

Yes, that sounds reasonable.  That's how diffutils does it (it doesn't
use readpipe).  However, it's a bit tricky to do:

* You have to quote the shell arguments correctly.  Diffutils uses
  quote_system_arg for that.

* You have to deal with pclose returning -1, as opposed to some
  other nonzero value.

* You have to deal with exit status 126 (not executable) and 127 (not
  found).  Hmm, I see now that Diffutils doesn't do that right -- I'll
  have to fix it....

readpipe isn't enough to do what Diffutils needs, e.g. it doesn't take
an arg vector.  I guess it could be generalized though.