gnuplot-beta

crash in write_history_n

[Bastian M. <bma...@we...>]

The version of write_history_n() using gnuplot's own version
of readline issues an int_error() when it cannot open the
history file for saving. As it is now called at program exit
(via atexit() or similar) it can no longer
bail_to_command_line(), which int_error() does. This causes
a segfault in that case. (Tested ;-) on a Win98 system without
%GNUPLOT% env. variable set.)
So this should be an int_warn(). The attached patch fixes that.

Greetings, Bastian

Re: crash in write_history_n

[Hans-Bernhard B. <br...@ph...>]

Bastian Maerkisch wrote:
> The version of write_history_n() using gnuplot's own version
> of readline issues an int_error() when it cannot open the
> history file for saving. As it is now called at program exit
> (via atexit() or similar) it can no longer
> bail_to_command_line(), which int_error() does. This causes
> a segfault in that case. (Tested ;-) on a Win98 system without
> %GNUPLOT% env. variable set.)
> So this should be an int_warn(). 

I'm not convinced that this is the right approach to the problem.
How about we make bail_to_command_line() a no-op if the command line is 
no longer there (i.e. test a global flag)?