Feat/patch management

[lukasrothenberger]

• adds patch generator
• adds patch applicator
• adds line_mapping initialization & bookkeeping

[goerlibe]

hardcoded clang and clang++ strings in CodeGenerator cause failure for me.
On my side I changed it to clang++-11 so I can continue on the GUI, but we should read it from our build_config

[goerlibe]

it would be helpful if the patch applicator would return an error code != 0 if applying the suggestions was not successful.
For now I will look if stdout contains "not successful."

[lukasrothenberger]

it would be helpful if the patch applicator would return an error code != 0 if applying the suggestions was not successful. For now I will look if stdout contains "not successful."

@goerlibe
what should be returned in case an already applied suggestion is to be applied?
0, since no issue arises, or 1 since it "can not" be applied?

fix proposed in 2378bd0

[lukasrothenberger]

hardcoded clang and clang++ strings in CodeGenerator cause failure for me. On my side I changed it to clang++-11 so I can continue on the GUI, but we should read it from our build_config

@goerlibe
proposed fix in c91eaca

[goerlibe]

it would be helpful if the patch applicator would return an error code != 0 if applying the suggestions was not successful. For now I will look if stdout contains "not successful."

@goerlibe what should be returned in case an already applied suggestion is to be applied? 0, since no issue arises, or 1 since it "can not" be applied?

fix proposed in 2378bd0

I am honestly fine with both. I think returning an error is more intuitive though, because the user tried to have something done and nothing was done.

Maybe we can use different error codes for different errors (e.g. 1=already_applied, 2=patch_failed, >=2 indicates a yet unknown error) and document them in the argparse interface.

Another (additional) option is to use stdout/stderr: I can show the output of stdout in an error message in the GUI and we can use stderr for the detailed error information (targeted at developers). This has the advantage that we do not need to update the GUI if we add more types of errors. Disadvantage is that we have to keep stdout clean.

[goerlibe]

rollback of simple_gpu resulted in a wrong line_mapping with the start_line being mapped to -1

@goerlibe
Edit: should be fixed by 8a11c29 and 8de3f0d

[goerlibe]

I can currently not apply more than one suggestion at once, not sure if I am doing something wrong, though

by "at once" I mean "after another", i.e. two consecutive calls to the discopop_patch_applicator with different ids

[goerlibe]

it would be nice if the patch_generator generates the applied_suggestions.json with an empty array.

@goerlibe:
Added in bdfaf62.

btw is there a reason for the entries being strings? they could be simple numbers, right?
@goerlibe: If I remember correctly, there was a compatibility reason, but I can not reproduce it in detail at the moment, sorry.