Hello all, I am involved in a project that has embedded ECL
(http://ecls.sourceforge.net/), a Lisp interpreter into Vim. Our end
goal for the project involves having Lisp code inside of Vim respond
to incoming data from a socket, so we also have a patch that allows us
to register callbacks that will be triggered when we receive data on a
file descriptor.
The code for this is here http://theclapp.org/repos/. The async
callback patch only really works for GTK and console I think.
Basically all the patch does is add a few more file descriptors to
poll/select on during Vim's read_char, if the appropriate descriptor
is triggered then we get a callback to Lisp.
The callbacks all appear to work fine, for example you can setup a
listening socket in Vim and have data output to a Vim buffer without
any problems. But there is a bug. Lets assume that we have a server
that simply echoes back data, and a client that sends data.
Data flow is something like:
(CLIENT) Embedded Lisp script sends data
(SERVER) Echoes data back
(CLIENT) Still sending data, there is also pending data coming back
from the server
(CLIENT) Finishes sending data, next time through the read_char loop
should trigger call back which will read data out from the socket.
However, the CLIENT side crashes in this scenario, it appears to crash
inside the Lisp interpreter. I suspect that somehow the last two
phases above are overlapping, and the interpreter is being re-entered,
which I don't think it supports. However, I don't see how this can be
happening - as far as I am aware Vim is totally single threaded (at
least *nix console builds), so the behaviour should be deterministic,
the data will finish sending, continue out to the main loop and then
trigger the callback.
I know it is a lot to ask for help on such a modified Vim, but this is
really starting to get me down. I would really appreciate any
thoughts about what might be going on, or how to debug this.
Cheers
Brad
