Re: pkg_add: Missing required library: /pkg/2024Q4/gcc10//lib/./libgcc_s.so.1

[Greg Troxel <gdt%lexort.com@localhost>]

Taylor R Campbell <riastradh%NetBSD.org@localhost> writes:

> Thanks for the details!  I thought I saw something fly by about a
> workaround for the fatal nature of this check, though, which was
> relatively new.  Currently this is probably breaking any bulk builds
> that use bootstrapped pkg_install rather than base pkg_install (which
> for all NetBSD releases predates the fatal check).

In pkgtools/pkg_install:

Index: files/add/perform.c
===================================================================
RCS file: /cvsroot/pkgsrc/pkgtools/pkg_install/files/add/perform.c,v
retrieving revision 1.128
diff -u -p -r1.128 perform.c
--- files/add/perform.c	7 Dec 2024 13:56:46 -0000	1.128
+++ files/add/perform.c	6 Jan 2025 13:14:37 -0000
@@ -1587,7 +1587,7 @@ pkg_do(const char *pkgpath, int mark_aut
 		if (check_dependencies(pkg))
 			goto nuke_pkgdb;
 
-		if (check_requires(pkg))
+		if (check_requires(pkg) && 0)
 			goto nuke_pkgdb;
 	} else {
 		/*
@@ -1598,7 +1598,7 @@ pkg_do(const char *pkgpath, int mark_aut
 		if (check_dependencies(pkg))
 			goto clean_memory;
 
-		if (check_requires(pkg))
+		if (check_requires(pkg) && 0)
 			goto clean_memory;
 
 		if (write_meta_data(pkg))


As I've said before, I believe this check should be backed out until it
doesn't cause false positives, and that it doesn't object to any package
that builds with PKG_DEVELOPER=yes.  (By false positive, I include
detecting something, that while wrong, is not the kind of integrity
problem the check is intended to catch.)

gcc10 being unusable breaks everything that needs fortran, even on
NetBSD-10.