Re: Strace

Hi Simon, On Wed, Jun 13, 2018 at 5:24 PM, Simon Peyton Jones wrote:

OK – so maybe the root cause is a framework failure – and indeed for the last few weeks I’ve seen

Framework failures:

plugins/plugins07.run plugins07 [normal] (pre_cmd failed: 2)

plugins/T10420.run T10420 [normal] (pre_cmd failed: 2)

plugins/T11244.run T11244 [normal] (pre_cmd failed: 2)

I have just learned to live with these failures, because I knew you were working on making things better. But it sounds as if they are still taking place.

The commit I made should have reduced the amount of failing tests to 0. framework failures are always quite unusual.

So:

- Yes, please make it not happen by default

I've removed the code, if you update it should be gone. It was there and

on by default because I was trying to debug failures on Harbormaster, I realized a switch isn't very useful as I won't be able to toggle it for Harbormaster anyway.

- - If you don’t get these framework failures, can we work together to resolve them?

These don't happen for me nor on Harbormaster, try picking a test, e.g T10420

run only that test to make sure it's not a threading issue: make TEST=T10420 test -C testsuite/tests If it still gives a framework error then do at the top level make VERBOSE=3 TEST=T10420 test -C testsuite/tests once it runs, the output should contain the command it ran as a pre_cmd, and the stdout and stderr from the pre_cmd output. Could you then send the error? if it doesn't show any of this, try make CLEANP=0 VERBOSE=3 TEST= T10420 test -C testsuite/tests --trace and copy and paste the pre_cmd command, which should just replay the action it did. Cheers, Tamar

Thanks

Simon

*From:* Phyx *Sent:* 13 June 2018 17:19 *To:* Simon Peyton Jones *Cc:* ghc-devs@haskell.org *Subject:* Re: Strace

Hi Simon,

The strace is only supposed to run when the normal test pre_cmd fails.

If it's running that often it means your tests are all failing during pre_cmd with a framework failure

https://git.haskell.org/ghc.git/blobdiff/4778cba1dbb6adf495930322d7f9e9 db0af60d8f..60fb2b2160aa16194b74262f4df8fad5af171b0f:/testsuite/driver/ testlib.py

But maybe I shouldn't turn this on my default. I'll pramaterize it when I get home.

Tamar.

On Wed, Jun 13, 2018, 17:09 Simon Peyton Jones wrote:

Tamar

I’m getting *megabytes* of output from ‘sh validate’ on windows. It looks like this

629 151745 [main] sh 2880 fhandler_base::fhaccess: returning 0

291 152036 [main] sh 2880 faccessat: returning 0

7757 159793 [main] sh 2880 fhandler_base_overlapped::wait_overlapped: wfres 0, wores 1, bytes 7

179457 1608947 [main] make 11484 fhandler_base_overlapped::wait_overlapped: wfres 0, wores 1, bytes 7

99 159892 [main] sh 2880 fhandler_base_overlapped::wait_overlapped: normal write, 7 bytes ispipe() 1

180 1609127 [main] make 11484 fhandler_base_overlapped::wait_overlapped: normal read, 7 bytes ispipe() 1

139 160031 [main] sh 2880 write: 7 = write(1, 0x6000396A0, 7)

142 1609269 [main] make 11484 fhandler_base::read: returning 7, binary mode

139 1609408 [main] make 11484 read: 7 = read(5, 0x60005B4B0, 7)

136 1609544 [main] make 11484 read: read(5, 0x60005B4B7, 193) blocking

4693 164724 [main] sh 2880 set_signal_mask: setmask 0, newmask 80000, mask_bits 0

but with hundreds of thousands of lines. (I have not counted)

I believe that it may be the result of this line, earlier in the log

cd "/c/Users/simonpj/AppData/Local/Temp/ghctest-8fa9s6rk/test spaces/./plugins/plugins07.run" && *strace* $MAKE -s --no-print-directory -C rule-defining-plugin package.plugins07 TOP=/c/code/HEAD/testsuite#

Note the strace.

That in turn was added in your commit

commit 60fb2b2160aa16194b74262f4df8fad5af171b0f

Author: Tamar Christina

Date: Mon May 28 19:34:11 2018 +0100

Clean up Windows testsuite failures

Summary:

Another round and attempt at getting these down to 0.

Could you perhaps have made a mistake here? Currently validate is unusable.

Thanks!

Simon