[GPGME PATCH] core: Restore get_max_fds optimization on Linux

Colin Watson cjwatson at debian.org
Tue Sep 19 21:36:53 CEST 2017


On Tue, Sep 19, 2017 at 06:53:56PM +0300, Alon Bar-Lev wrote:
> On 19 September 2017 at 18:42, Daniel Kahn Gillmor
> <dkg at fifthhorseman.net> wrote:
> > I support adoption of this patch in GPGME upstream.  It addresses a
> > pretty severe performance degradation in GPGME on Linux platforms that
> > was introduced between version 1.8.0 and 1.9.0.
> 
> Hi,
> 
> Just a question...
> What about of a solution of fork/exec a wrapper that closes handles
> then exec the actual binary?
> This wrapper may be considered as "trusted" and as it is exec-ed it
> can use any library call.

I did see that suggestion in earlier mailing list threads about this
problem, but it seemed quite a bit fiddlier than just doing it pre-exec
the way I did: building and installing a new program, sorting out paths
for it, and so forth looked likely to be more effort than (in this case,
and IMO) the relatively small amount of effort required to be
async-signal-safe.

It's obviously not my project and I don't object if somebody else wants
to put together a version involving an extra child process, but in all
honesty I'm not going to get round to it myself.

Thanks,

-- 
Colin Watson                                       [cjwatson at debian.org]



More information about the Gnupg-devel mailing list