On Tue, Mar 03, 2009 at 12:50:21PM -0500, Pierre A. Humblet wrote: > >----- Original Message ----- >From: "Christopher Faylor" <> >To: <cygwin-patches@cygwin.com> >Sent: Tuesday, March 03, 2009 10:38 AM >Subject: Re: [Patch] gethostbyname2 again > > >| On Mon, Mar 02, 2009 at 08:36:55PM -0500, Pierre A. Humblet wrote: >| > realloc_ent function, and call it from both dup_ent and the helper. >That >| > caused minor >| > changes in the 4 versions of dup_ent, and I don't know exactly what >| > format to use in the ChangeLog >| >| I would rather that you keep dup_ent as is so that there is no need to >| do run-time checks on the type. If you need to do something similar to >| what is currently in dupent, then couldn't you still create a >| realloc_ent but just pass in the destination pointer? Or even just make >| realloc_ent mimic realloc but do the rounding that seems to be the >| impetus for your breaking this out into a separate function. > >Chris, > >The impetus is that the new helper function is capable of formatting a fine >hostent in a single block of memory. So it doesn't need to have dup_ent >copy it in yet another memory block. > >However it still needs to store a pointer to the block of memory somewhere, >so that it can be freed later, and reusing the tls.locals seems logical. If it >does >that, then it must also free or realloc whatever is stored there. >That's what realloc_ent does. >The rounding is not essential, it's just nice to do it consistently in one >place. > >To avoid real-time checks, I could do as what dup_ent does, and have 4 versions >of the realloc_ent function, one main one with dst and sz arguments (that one >would be called by dup_ent without any run-time checks) and 3 (actually only >1 is needed for now) that invoke the main one with the correct dst based on the >type of the src argument . The src argument would be null but would have the >right type! That seems to meet your wishes. OK?
Yes. cgf
