[pacman-dev] portable package sorting for shell scripts

[Dave Reisner]

On Fri, Aug 05, 2011 at 09:19:43AM -0400, Dan McGee wrote:
> On Thu, Aug 4, 2011 at 9:01 AM, Dave Reisner <d at falconindy.com> wrote:
> > So it occurs to me that while it's all well and good to have a portable
> > package cache cleaner, it's no good at all if it all hinges on a GNU
> > sort flag (-V) which poorly approximates what vercmp does. Would we have
> > room for sort tool that implements alpm_pkg_vercmp? A simple
> > implementation could be as easy as the below... I think this would get a
> > lot of mileage out of the AUR/pacman wrapper crowd.
> >
> > thoughts?
> 
> Looks like this just takes a list of newline-separated versions on
> stdin then? Seems pretty reasonable, although argument support would
> be good as well. I'm also wondering whether supporting `pacman -Q`
> style input would be good as well (with package names and version
> numbers); thus it would be a smart sort going first by package name,
> but then interpreting the second component as a version rather than
> string.
> 
> > #include <alpm.h>
> > #include <stdio.h>
> > #include <stdlib.h>
> > #include <string.h>
> >
> > #define MAXLEN  4096
> If the tool is new, I'd prefer to avoid any buffer/line length restrictions.
> 
> >
> > int vercmp(const void *p1, const void *p2) {
> >  return alpm_pkg_vercmp(*(const char**)p1, *(const char**)p2);
> > }
> >
> > int main(void) {
> coding style, brace on next line.
> >  char **pkglist = NULL;
> >  char *pkg;
> This is a tad nitpicky, but since we never load actual package
> objects, might be wise to call these namelist and name instead.
> >  char buf[MAXLEN + 1];
> fgets reads at most n-1 characters and always null-terminates; no need
> for the extra byte.
> >  int i, pkgcount = 0;
> offsets should use size_t.
> >
> >  while (fgets(buf, MAXLEN + 1, stdin)) {
> If you heed my advice above the +1 will have to go away here.
> >    pkg = strndup(buf, MAXLEN);
> You already know buf is well-bounded, correct? strdup() is sufficient.
> >    *(pkg + strlen(pkg) - 1) = '\0';
> I'm going to throw a CRLF file at this to test. :)
> >    pkglist = realloc(pkglist, ++pkgcount * sizeof(*pkglist));
> Would be best to not reallocate every iteration, but instead do smart
> progression- start with something small/sane like 4, then double it
> each realloc. Also, the ++ works, but I'm not a fan of it inline like
> this.
> >    *(pkglist + pkgcount - 1) = pkg;
> If pkglist is a * instead of a **, I think you will be able to do something like
>     pkglist[pkgcount - 1] = pkg;
> instead. That also simplifies the dereferences needed in the printf()
> below as you can just do pkglist[i] instead. This is all untested, but
> I'm basing it roughly off what I ended up doing for the files array
> stuff in be_local.c.
> >  }
> >
> >  qsort(pkglist, pkgcount, sizeof(char*), vercmp);
> char(space)*, style rules.
> >
> >  for (i = 0; i < pkgcount; i++) {
> >    printf("%s\n", *(pkglist + i));
> >    free(*(pkglist + i));
> >  }
> >  if (pkgcount > 0) {
> >    free(pkglist);
> >  }
> >
> >  return 0;
> > }
> 

Of course. All lovely suggestions which will be taken into account. This
was just a proof of concept to get feedback before I spend any time on
it. Since it seems like I'm not crazy for thinking that this is
something useful, I'll clean it up to our standards.

d