GDB 7.7 release
lazarus / 12-Debugger

----------------------------------------------------------------------
Ocye · 2014-02-07 12:50

Quote from:
http://article.gmane.org/gmane.comp.gdb.announce/69Subject: GDB 7.7
released!
Newsgroups: gmane.comp.gdb.announce
Date: 2014-02-06 04:35:31 GMT

Release 7.7 of GDB, the GNU Debugger, is now available via anonymous
FTP.

I do not expect major improvements, or should I?

----------------------------------------------------------------------
Martin_fr · 2014-02-07 13:49

It will be a while  before I get to test it.

If you whish you can run the tests. debugger\test\Gdbmi\TestGdbmi.lpi

Note: You must recompile the IDE from the tools menu, before you
compile and run the test case.

Otherwise it will most likely complain about missing files. If it does
complain, ensure you delete all ppu files in
B:\lazarus_latest\debugger\test\Gdbmi\* and any subfolder.

This will give an indication if it broke anything between IDE and gdb.

Then of course, only long time tests can show, if gdb crashes or
crashes not. (7.6 crashed more often, than 7.5)

On windows32, there is no point in going for it, since
B:\lazarus_latest\debugger\test\Gdbmi\TestGdbmi.lpi is still open.

---
Otherwise question is what you expect to improve:

- properties: no, it does not. They first need to be possible in dwarf
(or maybe they are in dwarf 4 or 5 - not checked), thet in gdb, and
then fpc needs to add them to dwarf. That is a long way.

- calling functions from watches - same as properties

- string vs pchar: Not in dwarf2, but IIRC in dwarf 3 (gdb is
seriously unstable with dwarf3 (at least with dwarf3 encoded by fpc),
but that actually is subject to be tested in 7.7)

- speed: maybe, but not in the noticeable range 7.6 is also faster
than 7.2 but only if you add measuring code. Subjective it did not
feel like that, because if gdb improves by 10 percent, but then the
overall speed improves less than that, because time spent in the OS
and the IDE does not change.

----------------------------------------------------------------------
marcov · 2014-02-07 14:12

IIRC it listed x64 unwind support as feature.

----------------------------------------------------------------------
Ocye · 2014-02-11 18:35

Quote from: Martin_fr on February 07, 2014, 02:49:42 pmIf you whish
you can run the tests. debugger\test\Gdbmi\TestGdbmi.lpi
After your suggestion in this thread I was using 7.5, built from
source IIRC. So my first test was run with this version, second just
after installing gdb 7.7 from repos (Arch Linux). Lazarus from svn in
Qt flavor, fpc is 2.6.2.

Both tests report six errors, but the newer gdb has formatted output;
I didn't look at the details. Now I'm curious how it works in
productive mode.

Lazarus 1.3 r43996M FPC 2.6.2 x86_64-linux-qt

----------------------------------------------------------------------
Martin_fr · 2014-02-11 20:36

The tests are not perfect, so this sometimes happen.

the 7.5 test looks like there is an actual issue (the wrong memory was
shown for a watch that is set to as memdump of an object).
This may be because that needs some operation on hidden pointers
(object have a pointer) and gdb may apply the "^" deref differently


The 7.7 test shows that gdb did deref several pointers showing the
final result, rather than the pointer.
so unless you need to see the pointer, rather than the pointed too....

Also 7.7 sometimes reports records as class.... but the data is still
there.
probably https://sourceware.org/bugzilla/show_bug.cgi?id=16016