Incompatible bytecode error lacks file name of offending file

[moritz]

PackFile_unpack: This Parrot cannot read bytecode files with version 9.0.

PackFile header failed during unpack

This error message should include the name of the file that caused this error; otherwise it is very hard to figure out which file caused the problem.

(Happens on amd64 linux, with gcc 4.4.5)

[bacek]

Hello.

I think it's good "complex GCI task" candidate.

-- Bacek

[plobsing]

PackFile header failed during unpack

This error only occurs when an arbitrary string has been passed to ImageIO for thawing (and it has an invalid PackFile header). These strings are, in general, not related to files in any way.

This is made clear by the exception type thrown, EXCEPTION_INVALID_STRING_REPRESENTATION, clearly indicating that it is a string-related, not a file-related, error.

[dukeleto]

I am sure plobsing is correct, but this ticket should not get closed until the GCI task that mentions it is changed or deleted.

[nwellnhof]

This is probably triggered when linking against an incompatible libparrot and thawing the config hash in Parrot_gbl_set_config_hash_interpreter() in src/global_setup.c which happens pretty early in init_world(). plobsing noted on IRC:

"And the solution to making that error easier to debug would be (a) to have access to a C level backtrace or (b) for it to be possible to handle exceptions from c (so Parrot_gbl_set_config_hash_interpreter() could spit out something intelligent about the situation)."

As a quick work around, we could also directly test the version of the config_hash before thawing. But I can't find a way to read PARROT_PBC_MAJOR and PARROT_PBC_MINOR from libparrot.

[nwellnhof]

Commit d922b8a235 should give the following error message:

Version %d.%d of config hash is invalid, expected %d.%d. You're probably linking against an incompatible libparrot.

[whiteknight]

The format-checking logic was moved into a new function from where nwellnhof had originally put it, and the message was generalized a little bit to be used for all cases of loading in a PMC image through the API. Does this satisfy the requirements of the ticket, or do we need something more?