Re: [PATCH 00/30] Stop throwing exceptions from signal handlers

[Simon Marchi <simon dot marchi at ericsson dot com>]

On 16-03-18 03:18 PM, Pedro Alves wrote:
> Aka, get rid of immediate_quit.
> 
> Throwing exceptions from signal handlers is undefined C++.  So for C++
> conversion we need to replace the immediate_quit mechanism by
> something else.
> 
> This series:
> 
> - Replaces all immediate_quit uses with something else.
> 
> - Removes a few custom SIGINT handlers, centralizing more on the
>   "quit" flag and GDB's default SIGINT handler.
> 
> - Fixes several classical signal handler / mainline code races, using
>   the self-pipe trick.
> 
> - Introduces a "struct serial_event" abstraction to hide the self-pipe
>   trick -- on Windows we don't use pipes at all, and on Linux, we
>   could use eventfd (but I didn't do that, at least yet).
> 
> - Stumbles on a few bogus clear_quit_flag uses, and thus completely
>   eliminates it, for good measure.
> 
> - Decouples target_interrupt from all-stop/non-stop modes
> 
> - Fixes a few cases of indefinitely-delayed Ctrl-C handling.
> 
> - Fixes/incorrect handling of EINTR in ser-*.c backends
> 
> - Goes through the tree fixing all target_terminal_ours calls that
>   should be target_terminal_ours_for_output instead.  We only end up
>   with a few target_terminal_ours calls, at places were we need to
>   handle input.
> 
> - Plugs a few leaks along the way.
> 
> - Fixes a bug where you can't answer y/n to an internal error if the
>   target has the terminal.
> 
> - Introduces the concept of quit handlers.
> 
> - Does a few clean ups along the way.
> 
> Tested on x86-64 Fedora 23, native and gdbserver.
> 
> A slightly older version manually test on mingw64/Windows 7, to make
> sure Ctrl-C still works as expected there.
> 
> This applies on top of the "target mips" elimination patch:
>   https://sourceware.org/ml/gdb-patches/2016-03/msg00313.html
> 
> I've pushed the whole series to users/palves/immediate_quit.
> 
> Pedro Alves (30):
>   Don't rely on immediate_quit in command_line_input
>   Inline command_loop in read_command_line
>   TUI: check whether in secondary prompt instead of immediate_quit
>   Don't set immediate_quit in prompt_for_continue
>   Stop remote-fileio.c from throwing from SIGINT handler
>   Remove unused struct serial::name field
>   Introduce a serial interface for select'able events
>   Fix signal handler/event-loop races
>   Introduce interruptible_select
>   Make Python use a struct serial event
>   Don't call clear_quit_flag after check_quit_flag
>   Don't call clear_quit_flag in command_handler
>   Don't call clear_quit_flag in prepare_to_throw_exception
>   Don't call clear_quit_flag in captured_main
>   Eliminate clear_quit_flag
>   Decouple target_interrupt from all-stop/non-stop modes
>   Pass Ctrl-C to the target in target_terminal_inferior
>   Fix inconsistent handling of EINTR in ser-*.c backends
>   ada-lang.c: Introduce type_as_string and use it
>   Use target_terminal_ours_for_output in cp-support.c
>   Use target_terminal_ours_for_output in exceptions.c
>   Use target_terminal_ours_for_output in infcmd.c
>   Use target_terminal_ours_for_output in warning/internal_error
>   Add missing cleanups to defaulted_query and prompt_for_continue
>   Do target_terminal_ours in query & friends instead of in all callers
>   Use target_terminal_ours_for_output in MI
>   TUI: GC tui_target_has_run
>   target remote: Don't rely on immediate_quit (introduce quit handlers)
>   Eliminate immediate_quit
>   Eliminate target_check_pending_interrupt
> 
>  gdb/Makefile.in        |   6 +-
>  gdb/ada-lang.c         |  74 ++++++-----
>  gdb/cp-support.c       |   4 +-
>  gdb/defs.h             |  58 +++++++--
>  gdb/event-loop.c       |  41 ++++--
>  gdb/event-loop.h       |  26 +---
>  gdb/event-top.c        | 165 ++++++++++++++++++++++---
>  gdb/exceptions.c       |  17 ++-
>  gdb/extension-priv.h   |   5 +-
>  gdb/extension.c        |  34 +++--
>  gdb/gdb_select.h       |  15 +++
>  gdb/gnu-nat.c          |  12 +-
>  gdb/guile/guile.c      |   1 -
>  gdb/guile/scm-ports.c  |   4 +-
>  gdb/i386-tdep.c        |  49 ++------
>  gdb/infcmd.c           |  10 +-
>  gdb/linux-nat.c        |  10 --
>  gdb/linux-record.c     |  54 +++-----
>  gdb/main.c             |   1 -
>  gdb/mi/mi-interp.c     | 125 ++++++++++++++++---
>  gdb/mi/mi-main.c       |   7 +-
>  gdb/mingw-hdep.c       |  52 +-------
>  gdb/nto-procfs.c       |   4 -
>  gdb/posix-hdep.c       |  13 --
>  gdb/python/python.c    |  44 +++----
>  gdb/record-full.c      |  22 ++--
>  gdb/remote-fileio.c    |  34 +----
>  gdb/remote-sim.c       |   5 +-
>  gdb/remote.c           | 329 +++++++++++++++++++++----------------------------
>  gdb/ser-base.c         |  30 ++++-
>  gdb/ser-event.c        | 220 +++++++++++++++++++++++++++++++++
>  gdb/ser-event.h        |  51 ++++++++
>  gdb/ser-go32.c         |   4 +
>  gdb/ser-unix.c         | 108 ++--------------
>  gdb/serial.c           |  66 ++++++----
>  gdb/serial.h           |   8 +-
>  gdb/target-delegates.c |  25 ++--
>  gdb/target.c           |  17 ++-
>  gdb/target.h           |  19 +--
>  gdb/top.c              |  67 +++++-----
>  gdb/tui/tui-hooks.c    |  24 ----
>  gdb/tui/tui-io.c       |   2 +-
>  gdb/tui/tui.c          |   4 +-
>  gdb/ui-file.c          |   8 +-
>  gdb/utils.c            |  66 ++++++----
>  45 files changed, 1115 insertions(+), 825 deletions(-)
>  create mode 100644 gdb/ser-event.c
>  create mode 100644 gdb/ser-event.h
> 

I went over the series, and I understood some of the words used.  What I understood
looked very good.  For the rest, I trust you are doing the right thing. :)

Thanks a lot for doing this, since this is the major roadblock before we can't start
using C++!