PR python/17372 - Python hangs when displaying help()
authorPedro Alves <palves@redhat.com>
Thu, 23 Oct 2014 16:13:35 +0000 (17:13 +0100)
committerPedro Alves <palves@redhat.com>
Wed, 29 Oct 2014 17:29:26 +0000 (17:29 +0000)
commitd3d4baedb6d247c6372678edd15195a1a93c2c6c
tree7db1f5216dddae6dc675ee5291fc3ab3ea9ecaa1
parentd1e8523e40ed5094ed7d5b352ac6b0eabf9f690c
PR python/17372 - Python hangs when displaying help()

This is more of a readline/terminal issue than a Python one.

PR17372 is a regression in 7.8 caused by the fix for PR17072:

 commit 0017922d0292d8c374584f6100874580659c9973
 Author: Pedro Alves <palves@redhat.com>
 Date:   Mon Jul 14 19:55:32 2014 +0100

    Background execution + pagination aborts readline/gdb

    gdb_readline_wrapper_line removes the handler after a line is
    processed.  Usually, we'll end up re-displaying the prompt, and that
    reinstalls the handler.  But if the output is coming out of handling
    a stop event, we don't re-display the prompt, and nothing restores the
    handler.  So the next input wakes up the event loop and calls into
    readline, which aborts.
...
    gdb/
    2014-07-14  Pedro Alves  <palves@redhat.com>

        PR gdb/17072
        * top.c (gdb_readline_wrapper_line): Tweak comment.
        (gdb_readline_wrapper_cleanup): If readline is enabled, reinstall
        the input handler callback.

The problem is that installing the input handler callback also preps
the terminal, putting it in raw mode and with echo disabled, which is
bad if we're going to call a command that assumes cooked/canonical
mode, and echo enabled, like in the case of the PR, Python's
interactive shell.  Another example I came up with that doesn't depend
on Python is starting a subshell with "(gdb) shell /bin/sh" from a
multi-line command.  Tests covering both these examples are added.

The fix is to revert the original fix for PR gdb/17072, and instead
restore the callback handler after processing an asynchronous target
event.

Furthermore, calling rl_callback_handler_install when we already have
some input in readline's line buffer discards that input, which is
obviously a bad thing to do while the user is typing.  No specific
test is added for that, because I first tried calling it even if the
callback handler was still installed and that resulted in hundreds of
failures in the testsuite.

gdb/
2014-10-29  Pedro Alves  <palves@redhat.com>

PR python/17372
* event-top.c (change_line_handler): Call
gdb_rl_callback_handler_remove instead of
rl_callback_handler_remove.
(callback_handler_installed): New global.
(gdb_rl_callback_handler_remove, gdb_rl_callback_handler_install)
(gdb_rl_callback_handler_reinstall): New functions.
(display_gdb_prompt): Call gdb_rl_callback_handler_remove and
gdb_rl_callback_handler_install instead of
rl_callback_handler_remove and rl_callback_handler_install.
(gdb_disable_readline): Call gdb_rl_callback_handler_remove
instead of rl_callback_handler_remove.
* event-top.h (gdb_rl_callback_handler_remove)
(gdb_rl_callback_handler_install)
(gdb_rl_callback_handler_reinstall): New declarations.
* infrun.c (reinstall_readline_callback_handler_cleanup): New
cleanup function.
(fetch_inferior_event): Install it.
* top.c (gdb_readline_wrapper_line) Call
gdb_rl_callback_handler_remove instead of
rl_callback_handler_remove.
(gdb_readline_wrapper_cleanup): Don't call
rl_callback_handler_install.

gdb/testsuite/
2014-10-29  Pedro Alves  <palves@redhat.com>

PR python/17372
* gdb.python/python.exp: Test a multi-line command that spawns
interactive Python.
* gdb.base/multi-line-starts-subshell.exp: New file.
gdb/ChangeLog
gdb/event-top.c
gdb/event-top.h
gdb/infrun.c
gdb/testsuite/ChangeLog
gdb/testsuite/gdb.base/multi-line-starts-subshell.exp [new file with mode: 0644]
gdb/testsuite/gdb.python/python.exp
gdb/top.c