[Tarantool-patches] [PATCH] gdb: fix the extension to be loaded with Python 2

Kirill Yukhin kyukhin at tarantool.org
Wed Jul 22 15:46:56 MSK 2020


Hello,

On 04 июл 22:58, Igor Munkin wrote:
> There was a mystic error when the extension was loaded against old gdb
> versions build against Python 2:
> | (gdb) source luajit-gdb.py
> | Traceback (most recent call last):
> |   File "luajit-gdb.py", line 702, in <module>
> |     load(None)
> |   File "luajit-gdb.py", line 699, in load
> |     'lj-gc': LJGC,
> |   File "luajit-gdb.py", line 687, in init
> |     command(name)
> |   File "luajit-gdb.py", line 468, in __init__
> |     gdb.write('{} command initialized\n'.format(name))
> | ValueError: sequence.index(x): x not in sequence
> 
> I made a little investigation (for more info see the mentioned issue)
> and found the next fun fact: the exception was raised much earlier to
> <str.format>, more precisely in <gdb.events.new_objfile.disconnect>.
> However, the handled exception is preserved until <str.format> call and
> hits the condition underneath leading to the extension load failure.
> 
> As a result to avoid the exception raise, the special global variable is
> introduced for legacy (i.e. Python 2) environment. It checks whether any
> callback is associated with new_objfile event prior to disconnecting it.
> This variable usage is encapsulated within two introduced routines:
> <connect> and <disconnect> which are wrappers for ones provided by gdb.
> 
> Furthermore, after diving to gdb sources related to Python embedding, I
> found that callbacks are grouped into an internal list. Previous
> implementation appended the <load> function to this callback list on
> each its unsuccessful call, but only the successful one is removes it
> from the list. Thereby disconnect action is moved prior to connect one
> so there is no more than one <load> instance kept in callback list.
> 
> Fixes tarantool/tarantool#4828
> 
> Reported-by: Oleg Babin <olegrok at tarantool.org>
> Signed-off-by: Igor Munkin <imun at tarantool.org>

I've checked your patch into Tarantool's luajit repo and bumped a
new version in 1.10, 2.4, 2.5 and master branches.

--
Regards, Kirill Yukhin


More information about the Tarantool-patches mailing list