Re: [PATCH] PR mi/2086 -break-insert missing error diagnostic

[Nick Roberts <nickrob at snap dot net dot nz>]

 > > This patch fixes PR mi/2086.  There may be similar bugs arising from using
 > > catch_exceptions_with_msg inappropriately.
 > 
 > Is this the whole patch?  I guess you updated the caller too.

I have other changes but they're all older.  I think that's it but I'll
check more carefully.

 > I think removing catch_exceptions_with_msg isn't enough - if you look
 > at the call site, it's not prepared for an error() to escape there.
 > It would have to use a cleanup for deprecated_set_gdb_event_hooks.

Heres my take:

I think the problem arises from changes Andrew Cagney made on 2005-01-13.
It worked before then but was probably a bit of a hack.  By removing
catch_exceptions_with_msg the error propagates back to catch_exception
with the function argument captured_mi_execute_command:

#14 0x0812fdf8 in catch_exception (uiout=0x8823918, func=0x80ca1e7 <captured_mi_execute_command>, func_args=0xbfc229a4, mask=6) at exceptions.c:469
#15 0x080ca5de in mi_execute_command (cmd=0x8855220 "-break-insert *", from_tty=

(mi_execute_command was also changed on 2005-01-13).

 > Ulrich posted a detailed analysis of this to gdb@ on Oct. 4th.  I think
 > it was just a buggy conversion to catch_exceptions_with_msg; we should
 > check everywhere touched by whatever patch that was.  Looks like it was
 > from 2005-01-13.

I'll take a look but I'm guessing 2086 might be less general.

 > > I could slowly work my way through the mi PRs in the bug database.  Can I
 > > close a bug or is that another level? (I don't know my password, if I have
 > > one).
 > 
 > I'll send it offlist.  Feel free to close bugs.

Thanks.

-- 
Nick                                           http://www.inet.net.nz/~nickrob