On Wed, Feb 3, 2010 at 4:08 PM, Dominique Toupin
<dominique.toupin@xxxxxxxxxxxx>
wrote:
It does give the
impression you're selective about which communities are worth encouraging (EDC)
and which to do an end run around
(GDB).
Given that I have no vested interest in GDB, I'm not sure why I
personally need to promote it. EDC is an interesting addition to our ecosystem
and I'm curious to see where it fits in. And I'm especially curious if it
helps solve my JNI debugging problem.
My understanding
from CDT meeting discussions (which don't have minutes most of the
time) was that we would leave
CDI for Galileo and put DSF-GDB for Helios, in
the last meeting (which has documented minutes), the plan for CDT
is not clear anymore ("Which one should be the default? EDC is good for Windows"). Not having a clear
direction doesn't help companies to contribute to open
source projects and doesn't help to focus the existing community, it
takes time to be up to speed with a new framework, people are hesitant
to do it unless they know the knowledge will help them in the future and
that a stable/rich set of features is
available.
The direction is decided by the community and what people contribute
to. Pretty much all current contributions are based on DSF. So if there is
confusion over whether to pick DSF or CDI, then it should be clear. CDI is a
deprecated framework.
Now, whether people pick EDC over GDB, is a
matter for the community and we are giving them the choice. I have no idea
what the people doing the 400,000+ downloads of CDT prefer. Today they are
using CDI with GDB. So if we're going to switch to whichever, we need to make
sure we don't mess them up too badly. Other than that, the adopting vendors
and distribution makers can pick whatever debugger and framework they
like.
In
the debug area, many companies are making big contributions to
GDB, e.g. multi-architecture, multi-OS support (IBM,
Ericsson), Python scripting (RedHat, etc), dynamic and static Tracepoint
(CodeSourcery, Ericsson, Polytechnique), improved C/C++ debug
(RedHat/Archer), Multi-Process/exec (CodeSourcery, Freescale, etc.),
non-stop (CodeSourcery, Ericsson) reversible debug (VMware, Virtutech,
etc.), I could go on, especially for upcoming features like core awareness.
A lot more companies are making smaller contribution and using GDB
extensively, DSF/GDB has a lot of features, it's been around
for many years, GDB stubs works on all targets that I know of,
from RTOS, to native Linux, to windows, to
emulator/simulators (e.g simics), to JTAG, we can even use the
same DSF/GDB binary on host to debug different targets. GDB is the
most ubiquitous debugger I know. For the DSF-GDB, I think
WindRiver (e.g. Pawel, Randy, etc.) did a great job with the initial
DSF-GDB implementation and we were happy to contribute to it in order to
help build the future debug solution for
CDT.
Again, GDB is just one of the external debuggers that we integrate
with. It's the most popular, but it's not up to the CDT to pick what debugger
people use. CDT is a platform. Don't confuse the exemplary integration we have
with the gnu tools as an endorsement of them, only an acknowledgement of their
proliferation. The fact that we have integrations with the gnu tools actually
enters a grey zone with the IP rules at Eclipse, so we do need to tread
carefully.
I am afraid that if
CDT changes direction again, companies who are using GDB and
Eclipse will continue to wait for contributions until CDT makes up
it's mind and has a stable/rich set of features for debug,
going EDC as
default is narrowing and doesn't help collaboration with
Linux Foundation on tools (something equivalent to LSB but for tools).
As you know we are planning a packaging of CDT + Linux Tools + GNU
Tools for Linux targets, hopefully it will give more visibility to the
CDT in the Linux community and attract more people (Linux is a fast
growing community). We can do it on Eclipse Lab under some form of CDT
distribution including GNU
binaries.
That's fine. As you do your own distribution, you are free to pick
the debugger and integration you'd like. The CDT is a platform that supports
many debugger integrations. You aren't forced into any particular
one.
To
answer your question below yes DSF/GDB is working with MinGW gdb 7.0.1 on
windows and you can use it in Wascana, hopefully we can work
together on the CDT
distribution.
Wascana is a pet project of mine, for legal reasons not a part of the
CDT project, but intended to help promote CDT in the Windows development
community. And, as it is turning out, it's an exemplary implementation of an
p2 installer that gives you a complete IDE including tools and libraries. Feel
free to reuse some of the concepts from it and the cdt.p2 plug-in and I can
help you get started with that.
For Linux, I'd prefer to spend time
helping the Linux distro guys who all have their own external tools and
install strategy. Fedora is doing a great job at integrating the CDT. Ubuntu
and Debian need help. I might do something with Moblin, but I'd prefer they
release an SDK with CDT in it. And Android is something different all
together. That's enough to fill my days.
I guess putting something
together in order to make the DSF-GDB default choice more obvious would be
good, the above + list of features could be a
start?
Sure. However, my plan for the next couple of weeks is to take
another look at our launch story. I hate the fact we have a default. It should
be related to the toolchain that the user is using for their project. Then
when you define your toolchain, you can say what debugger and integration
framework to use. For our exemplary toolchain support, where DSF/GDB works
well, we can hook it up. Where it doesn't we can use alternatives.
I
need to and plan on being non-partisan in this debate of GDB versus EDC. I
hope you aren't looking to me for guidance on what to choose. That's for you
the community to decide, if you feel you have to at
all.
Doug.
Well, I certainly wasn't making a personal attack on the people
working on gdb for mingw, nor was I challenging their character. The
issues I was running into was much more technical than that and involved
stack unwinding when mixing msvc and gcc built DLLs. Or maybe it was
stopping on shared library events giving me garbage stacks or stopping at
the wrong time. I'm not a debugger expert, but It was more complicated
than tty's at any rate.
I'm just not confident that the Windows
platform is getting a lot of love from the gcc and gdb communities, and if
that's the case, I'm OK with that.. It is a bit of a niche market. Or
maybe that's changing lately and I'll certainly take a look at the 7.0.1
gdb that's now up on the mingw site. If it works with GDB/DSF, then I'll
be happy. If not, I do know that EDC seems to work there and there is
effort going into making it work well
there.
Cheers,
Doug.
On Wed, Feb 3, 2010 at 9:18 AM, Daniel Jacobowitz
<dan@xxxxxxxxxxxxxxxx> wrote:
This
really pushes my buttons.
Instead of constantly complaining about
MinGW GDB, you could submit
bug reports or try to understand the
actual problem. For someone
who's so gung-ho about community
involvement, you're selective about
which communities are worth being
involved in (or even worth
encouraging) and which to do an end run
around.
For instance, this might be the mostly intractable
problem of
_isatty(). There's no API equivalent to pseudo-TTY's
on Windows;
if you run a console program from GDB, it has to either
pop up a new
console or else run the program with its standard
handles connected
to a pipe. This results in buffered output
and people complain about
missing printfs. I think current GDB
does a little better.
An experienced Windows developer might be able
to make it do
better still.
It could be any number of other
problems. This random poster is
complaining about GDB 5.3.
That is a seven year old release. For
some perspective on
how long ago that is: this is like walking up to
you and complaining
about a bug in CDT 1.0.
--
Daniel
Jacobowitz
CodeSourcery
_______________________________________________
cdt-dev
mailing list
cdt-dev@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/cdt-dev
_______________________________________________
cdt-dev
mailing list
cdt-dev@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/cdt-dev
