FYI: fix Solaris problem with auto host charset

FYI: fix Solaris problem with auto host charset

[Tom Tromey]

I'm checking this in.

On some versions of Solaris, nl_langinfo(CODESET) will return "646"
for the C locale.  Unfortunately, this value is not then recognized by
iconv_open.

This fixes the problem by recognizing this situation and using "ASCII"
instead.  I'm unaware of any other names needing this treatment.

This is a bit of a hack.  But, I think such cases should be rare.  (My
experience with gcj was that this was the only such case ever
reported.)

Built and regtested on x86-64 (compile farm).
Joel also tested it on an actual Solaris box showing the problem, and
reported improved test results.

Tom

2009-04-17  Tom Tromey  <tromey@redhat.com>

	* charset.c (_initialize_charset): Add special case for "646".

Index: charset.c
===================================================================
RCS file: /cvs/src/src/gdb/charset.c,v
retrieving revision 1.22
diff -u -r1.22 charset.c
--- charset.c	15 Apr 2009 22:20:30 -0000	1.22
+++ charset.c	17 Apr 2009 23:45:32 -0000
@@ -817,6 +817,10 @@
 #ifndef PHONY_ICONV
 #ifdef HAVE_LANGINFO_CODESET
   auto_host_charset_name = nl_langinfo (CODESET);
+  /* Solaris will return `646' here -- but the Solaris iconv then
+     does not accept this.  */
+  if (!strcmp (auto_host_charset_name, "646"))
+    auto_host_charset_name = "ASCII";
   target_charset_name = auto_host_charset_name;
 
   set_be_le_names ();

Re: FYI: fix Solaris problem with auto host charset

[Eli Zaretskii]

> From: Tom Tromey <tromey@redhat.com>
> Date: Fri, 17 Apr 2009 17:49:58 -0600
> 
> On some versions of Solaris, nl_langinfo(CODESET) will return "646"
> for the C locale.  Unfortunately, this value is not then recognized by
> iconv_open.
> 
> This fixes the problem by recognizing this situation and using "ASCII"
> instead.  I'm unaware of any other names needing this treatment.

Btw, why is the code assuming that nl_langinfo always returns a
meaningful result?  It can potentially return an empty string.
Although this is at worst a very rare situation, perhaps we should
take it into consideration and fall back on GDB_DEFAULT_HOST_CHARSET.

Also, I don't quite get the reason why nl_langinfo is not called in
the PHONY_ICONV case.  It seems we blindly assume the native charset
is Latin-1 in that case, but perhaps nl_langinfo, if it exists, will
give a better guess?

Re: FYI: fix Solaris problem with auto host charset

[Tom Tromey]

>>>>> "Eli" == Eli Zaretskii <eliz@gnu.org> writes:

Eli> Btw, why is the code assuming that nl_langinfo always returns a
Eli> meaningful result?

Oversight.

Eli> Also, I don't quite get the reason why nl_langinfo is not called in
Eli> the PHONY_ICONV case.  It seems we blindly assume the native charset
Eli> is Latin-1 in that case, but perhaps nl_langinfo, if it exists, will
Eli> give a better guess?

In the PHONY_ICONV case we have no way to convert between character
sets.  I think this only affects u'' and U''; so it really a matter of
which failure you want in this situation.

If you want to change this, it is fine by me.

Tom

Re: FYI: fix Solaris problem with auto host charset

[Tom Tromey]

>>>>> "Tom" == Tom Tromey <tromey@redhat.com> writes:

Eli> Btw, why is the code assuming that nl_langinfo always returns a
Eli> meaningful result?

Tom> Oversight.

I re-read the relevant standard text online, and as far as I can tell,
nl_langinfo will only return an empty string if the argument is invalid.
So, I don't think this is an issue for gdb.

Tom

Re: FYI: fix Solaris problem with auto host charset

[Eli Zaretskii]

> Cc: gdb-patches@sourceware.org
> From: Tom Tromey <tromey@redhat.com>
> Date: Tue, 21 Apr 2009 12:53:26 -0600
> 
> >>>>> "Tom" == Tom Tromey <tromey@redhat.com> writes:
> 
> Eli> Btw, why is the code assuming that nl_langinfo always returns a
> Eli> meaningful result?
> 
> Tom> Oversight.
> 
> I re-read the relevant standard text online, and as far as I can tell,
> nl_langinfo will only return an empty string if the argument is invalid.
> So, I don't think this is an issue for gdb.

I thought "invalid argument" includes arguments that are unsupported.
Doesn't it?

Re: FYI: fix Solaris problem with auto host charset

[Tom Tromey]

>>>>> "Eli" == Eli Zaretskii <eliz@gnu.org> writes:

Eli> I thought "invalid argument" includes arguments that are unsupported.
Eli> Doesn't it?

Not according to the text I read.

Tom