Parse failure for specific log file

[jaspervdj]

File and error reported here: jaspervdj/profiteur#22

I reproduced it using ghc-prof version 1.4.0.4

[maoe]

Looking at the prof file in jaspervdj/profiteur#22, it seems to be broken:

 34157      ==                                                                                                                                            Process                               src/Process.hs:22:54-55                                   25284          0    0.0    0.0     0.0    0.0
 34158       ==                                                                                                                                           Data.Text                             Data/Text.hs:(325,5)-(327,30)                             25285          2    0.0    0.0     0.0    0.0
 34159        aBA                                                                                                                                         Data.Text.Array                       Data/Text/Array.hs:84:7-9                                 25295          4    0.0    0.0     0.0    0.0
 34160                                                                                                Data.ASN1.Parse                       Data/ASN1/Parse.hs:34:27-30                               46923        306    0.0    0.0     0.0    0.0                                                    
 34161                                    return                                                                                                          Data.ASN1.Parse                       Data/ASN1/Parse.hs:48:5-24                                46962        153    0.0    0.0     0.0    0.0
 34162                                     pure                                                                                                           Data.ASN1.Parse                       Data/ASN1/Parse.hs:39:5-35                                46963        153    0.0    0.0     0.0    0.0

Look at the line 34160. There is no cost centre name in that line. Also return at the next line is indented too much. I'm not sure whether this is a bug in the GHC profiler or something was messed up when uploading the file. @varosi Any ideas?

[maoe]

BTW, maybe I should switch the ghc-prof parser from attoparsec to (mega)parsec or something for better error messages. Do you have any opinions to pick a parser library?

[varosi]

It's coming from a ZIP file. It's the same as on my machine.
File is generated with runing few concurrent processes with this RTS options:

+RTS -N -s -h -i0.1 -p -M1G -S/var/log/processJiraS.log -po/var/log/processJira.log

with GHC 8.2.2 (LTS 10.3)

[varosi]

@maoe about parser library, I don't have opinions.
But here is some list:
https://github.com/Gabriel439/post-rfc/blob/master/sotu.md

[maoe]

@varosi Can you check if you can generate a broken .prof file again? If it's reproducible, I guess we should report it to GHC trac.

In any case I'll probably give megaparsec a try for better error messages when I have time.

[varosi]

Here is another one, but it seems that is opening fine:
processJira.log.prof.zip

I think that if multiple processes try to create the same file - it could mess the file. In the perfect case they should wait one another or just use different (unique) file names and check if the file already exists.

[maoe]

@varosi Thanks for checking. In that case you might want to use the -po option to distinguish processes.

[varosi]

Thanks, but it is the same executable ran multiple times, i.e. all the processes have the same name, but different process id.

[maoe]

Yeah I understand that. I mean, can you specify a unique identifier for each process using -po option at runtime?

[varosi]

It's possible with redirecting via shell script, because currently the process is called from a application that we don't have control on.