# control bcc-ed tag 264887 + moreinfo thanks On Tue, Aug 10, 2004 at 02:41:27PM -0500, Jeff Licquia wrote: > Package: libc6 > Version: 2.3.2.ds1-13 > Tags: patch > > When the GLOB_APPEND flag is passed to glob(), the call is supposed to > tack new results on to the ones already represented in the glob_t passed > in. If there is an error, partial results may be added, but the > original glob_t should definitely be left alone. > > This is not respected in at least one part of glibc's glob > implementation. In this particular place, globfree() is called on the > glob_t regardless of the flags passed. Since glob() is implemented > recursively with GLOB_APPEND added to the flags, this can affect the > results returned even from a call where GLOB_APPEND is not set. > > The following patch fixes the one instance of this behavior I found. I've looked at the code, and if I do understand what you mean with your patch, I don't undestand how to trigger a case where the code would not work, please consider that your patch is applied in a part of the code that is very specific, and guarded by many checks, are you sure it isn't safe to proceed the way it's done ? Please provide a testcase here that your patch solves, without that the patch won't be applied as afaict it could be a memory leak as well. -- ·O· Pierre Habouzit ··O madcoder@debian.org OOO http://www.madism.org
Attachment:
pgpFeGIim7ht7.pgp
Description: PGP signature