Re: ANN: ncurses-6.0-20170624

[Allen Hewes]

On 6/25/2017 19:03, Thomas Dickey wrote:
> On Sun, Jun 25, 2017 at 08:50:01PM +0000, Allen Hewes wrote:
>> On 6/25/2017 6:02, Thomas Dickey wrote:
>>> On Sun, Jun 25, 2017 at 01:28:42AM +0000, Thomas Dickey wrote:
>>> ...
>>>> 20170624
>>>>    + modify c++/etip.h.in to accommodate deprecation of throw() and
>>>>      throws() in c++17 (prompted by patch by Romain Geissler).
>>>>    + remove some incorrect calls for memory-leaks from 20170617 changes
>>>>      (report by Allen Hewes).
>>>
>>> This should get the set of changes back on track.  I noticed some
>>> problem with the c++ demo (which wouldn't affect gdb, but might have
>>> a common root cause).
>>>
>>
>> Hi Thomas,
>>
>> Thank you for looking into MinGW issues. I realize it's not everyone's
>> cup of tea. There's still multiple free() happening.
> 
> I see.  Debugging gdb+readline+ncurses may seem a little unpalatable.
> 
> I've been focusing on recasting the problem into one that I can test with
> simpler programs.  But I overlooked the possibility of readline calling
> tgetent, e.g., before initscr.
> 
> Actually, just reading the configure options in your previous mail, I didn't
> see (yet) a significant difference between your/my configurations.  But today 
> I
> added a test-script with your settings, and given TERM=unknown, can see a
> problem to investigate (perhaps the one that your traceback/report shows).
> 

Setting both TERM= (unsets) and TERM=win32console in a Command Prompt
produces the same results.

I would guess that TERM=unknown is the same as TERM=dumb based on the
path I've traced. Any value that isn't found in terminfo would probably
produce the same results on Windows.

I think the tgetent behavior is because when you build gdb with TUI
support, it can still run in non-TUI mode. e.g. $ gdb (normal readline
mode), gdb --tui (ncurses mode). They use to build two binaries, one for
normal readline (gdb) mode and one for TUI (gdbt) mode but they stopped
doing that some time ago.

I am fiddling around with some local patches that remove
free()/_nc_free_termtype2() calls where the data structure was passed
into the function as reference (but not allocated therein). I'm not
normally a C hacker... so not sure what I will find (if anything at all).

Thanks again,

/allen

signature.asc
Description: OpenPGP digital signature