Le lundi 31 juillet 2006 à 16:00 +0200, Frederic RISS a écrit :
>  I'll follow up soon with a patch doing this.

Here's a code patch for discussion and a tentative doco patch. I've
chosen to initialize the variable early in handle_inferior_event, so
that it's set whatever happens. I thought that if we document it, it'd
rather be always set and not only for breakpoint condition evaluation. 

I'm still wondering about 2 things:

1. Should it change on a user requested thread switch? That would be a
simple patch to thread.c:switch_to_thread. But maybe documenting it as
'the id of the last stopped thread' wouldn't be a bad idea. I don't
think GDB provides a way for the user to get this information (I mean
once he has selected another thread). This way 'thread $_gdb_thread'
would always bring you back to the stopped thread.

2. Should we add support for read-only convenience variables? Writing in
$_gdb_thread makes no sense... but it causes no harm either. We already
have $bpnum in the same situation. I see some ways to support this:
        - Add lval_internalvar_readonly 
        - Add a read_only flag to struct internalvar and check it at
        assignment time. Internal vars are already special-cased there,
        so this should be easy. Would be my solution of choice.
        - Don't create a real convenience variable, but add OP_THREADID
        to expression.h and generate it in write_dollar_var. Also add
        some API to infrun.c to export the stopped threadid.

Fred.