Re: pkg/42256: check-files with /usr/pkgsrc mounted read-only

[Steven Drake <sdrake%xnet.co.nz@localhost>]

The following reply was made to PR pkg/42256; it has been noted by GNATS.

From: Steven Drake <sdrake%xnet.co.nz@localhost>
To: gnats-bugs%NetBSD.org@localhost
Cc: 
Subject: Re: pkg/42256: check-files with /usr/pkgsrc mounted read-only
Date: Tue, 3 Nov 2009 20:37:11 +1300 (NZDT)

 On Mon, 2 Nov 2009, Joerg Sonnenberger wrote:
 
 >On Mon, Nov 02, 2009 at 12:40:02AM +0000, Steven Drake wrote:
 > > Your suggesting that rm -f treats unlink(2) failures that return an EROFS 
 > > errno value specially, and by looking at the source's nether the coreutils
 > > rm or the netbsd rm do.
 >  
 > Yes, but the files shouldn't exist in first place. They are phony
 > targets.
 
 I remembered that there phony after I sent that email (Oops!).
 
 Anyway I spent some more time looking at netbsd/coreutils rm sources (and
 debugging the coreutils rm).
 
 The netbsd rm does any fts_open/fts_read on the file and look at fts_info,
 form the source:
 /*
  * FTS_NS: assume that if can't stat the file, it
  * can't be unlinked.
  */
 
 The coreutils rm does an unlinkat and checks the errno value on failure
 ignoring it if its ENOENT.  With linux/glibc (at least) unlinkat/unlink
 sets errno to EROFS for a non-existent file on a read-only filesystem.
 
 That begs the question, what should unlink set errno to?  I like to know
 what other OSes/libcs do?
 
 In any case I'll submit this as a coreutils bug, (who know what it's 
 maintainers will think?), but it is probably easier to apply the patch I 
 sent.
 
 -- 
 Steven