Parametric macro arguments are not expanded

[ascherer]

There seems to be a timing problem when expanding

%bcond_without patches
...
%prep
%autosetup -c %{?with_patches:-p1}%{!?with_patches:-N}

This was my original version in sgb.spec (developed with debbuild) where the conditional macro %{?with_patches:-p1} does not produce the expected -p1 option for %autosetup, independent of the use of --without patches on the CLI.

Using classic %if..%else..%endif logic works, but this takes five LOC and I'd rather like the single-line solution.

This occurs with rpmbuild 4.12.0.1 on Kubuntu 16.04.

As an aside I report that adding BuildArch: x86_64 to sgb.spec makes rpmbuild ignore the patches completely.

[pmatilai]

Yup, see also https://bugzilla.redhat.com/show_bug.cgi?id=1397209. Whether it's truly intentional I don't know but AFAIK it's always been that way.

Fixing it is a simple matter of:

--- a/rpmio/macro.c
+++ b/rpmio/macro.c
@@ -687,8 +687,10 @@ grabArgs(MacroBuf mb, const rpmMacroEntry me, const char * se,
      * Append the results to main argv. 
      */
     {  ARGV_t av = NULL;
-       char *s = xcalloc((lastc-se)+1, sizeof(*s));
-       memcpy(s, se, (lastc-se));
+       char *s = NULL;
+
+       /* Expand possible macros in the arguments */
+       expandThis(mb, se, lastc-se, &s);
 
        argvSplit(&av, s, " \t");
        argvAppend(&argv, av);

...but it'd be changing a longstanding behavior and there's no telling what might break if we did that.

[pmatilai]

Looking at Fedora rawhide specs, at least font packages break with the above change. Pretty much as expected, actually.

[ascherer]

So it seems I went a bit overboard with my implementation of %autosetup in debbuild. Interesting though, that the same issue occurred on RH. 😉

[pmatilai]

I didn't really mean to, but ended up looking at this a bit more.

It's by no means just the font-packages that break, there are all sorts of other cases too, at least:

• passing arguments starting with "-" by passing them in as macros (%py2_build etc macros, some java packages) which now fail as invalid options. This is basically abuse, using "--" to separate arguments from options actually works but isn't widely known or documented.
• %systemd_user_post macro fails with recursion loop on %{?*}

Breakage on automatic macros is not a big surprise considering this changes scoping rules in some ways.

[ascherer]

In the particular case of %autosetup in sgb.spec, using -N or -p1 depending on the CLI option --without patches is utterly redundant, because the set of patches is commented-out in the first place.

[ignatenkobrain]

workaround which I use in %py3_build is %py3_build "--use-cython" which works....

I would reopen this bug.

[pmatilai]

Yes the issue is valid regardless of the particular sgb.spec case, reopening.

[pmatilai]

The case in RH bugzilla kind of highlights the insanity nicely. The first one works, the second does not:

%setup -q %{?rev:-c}
%autosetup %{?rev:-c}

Yes, one is a builtin and the other one is a macro. But I'd be hard-pressed to come up with a sensible explanation why those should behave differently.