Heads up! Working on bugpoint

I'm starting work on getting bugpoint working with remote execution for arm. Is anyone else working on this?

David

Hello David,

I'm working on bugpoint.

There are 2 patched out there for review that fix problems and make remote execution work.
I'm looking at adding it to the test-suit and use make to run tests remotely for cross-compilation (following Evan's suggestion).

Let's coorddinate our effort.

Best regards,
Viktor

Hi Viktor,

I did not see your patches, can you send them again. Here are the changes that I have made that seem to get remote execution working for bugpoint. I have not yet changed the test system to provide -remote-client and -remote-port.

David

patch2 (1.86 KB)

David,

Please find attached 2 diffs. Please notice that RemoteRunSafely.sh may get moved to the test suite sub-tree.

We have targeted different areas fso far which is great.
What's your plan?

I have been thinking of using -run-custom and -safe-run-custom argumants for the cross-tools to keep bugpoint out of knowing any target details ans such.

Best regards,
Viktor

makeAbsolute.patch (2.89 KB)

ToolRunner.diff (4.13 KB)

I just checked in my changes to bugpoint and Makefile.programs in the test suite. These changes shouldn't conflict with yours.

revision 75292
revision 75293

bugpoint now works for remote ARM if you run it directly from the command line. It doesn't work correctly when invoked from the test makefile because envvar PWD is being set to /bin/pwd and so causing GCC::ExecuteProgram to do the wrong thing. I'm not sure what is going on with that, perhaps something in the makefile hierarchy is setting PWD to /bin/pwd instead of executing that command to get the pwd. I will investigate.

David

I will investigate.

I'm in the middle of it as well.

Best regards,
Viktor

I fixed the PWD weirdness. bugpoint now works with remote ARM from testsuite makefiles.

Sending External/SPEC/Makefile.spec
Sending Makefile.programs
Transmitting file data ..
Committed revision 75307.

David

Hello, David

bugpoint now works for remote ARM if you run it directly from the
command line. It doesn't work correctly when invoked from the test
makefile because envvar PWD is being set to /bin/pwd and so causing
GCC::ExecuteProgram to do the wrong thing. I'm not sure what is going
on with that, perhaps something in the makefile hierarchy is setting
PWD to /bin/pwd instead of executing that command to get the pwd. I
will investigate.

This might be propagated from Makefile.config. I think it's better to
figure out why it's propagated from there instead of hardcoding paths
into testsuite. This might break mingw32, for example.

I don't think it is being propagated from anywhere. I don't thing you can rely on PWD being defined in make like you do from a shell. Other places in the makefile also explicitly need to calculate PWD using $(shell pwd) so I don't think this usage is unusual. But looking at it some more it does seem that I might be able to use an already available variable that is calculating pwd. I give it a try.

David