Format trace to JSON format

[denismerigoux]

Based on @rprimet's, excellent suggestion, we want to print the trace by the interpreter in a JSON format that could be exploitable by debug tools. However, the parsing of raw events to events is broken now and needs to be fixed. In the meantime, this PR prints the list of raw events on the standard output in JSON format.

[rprimet]

Note : it is possible that we are not outputting valid json yet (for instance, numbers are a footgun in json, and it seems that numbers with an unset fractional part are not accepted : e.g. 1.0 is valid but 1. isn't -- that should probably be an easy fix)

[AltGr]

LGTM! An area that could be improved is the UI: we have 3 different flags for traces, and specifying an output file for example doesn't imply that the trace should be on.
Ideas:

• make the trace output file be an optional argument of --trace, so that no extra flag is needed (note though: it could break existing scripts that do e.g. catala interpret -t foo.catala_en because foo.catala_en would be taken as the argument of -t ; we don't have that many scripts around and I still think it's ok, though)
• add environment variables for the flags (in particular, that could be useful for the trace format)
• infer the trace format from the output file name if .json

This may require some habit of Cmdliner so I can help ^^

[rprimet]

Adding the output file as an option of the --trace argument appears to have broken some scripts ( https://github.com/CatalaLang/catala/actions/runs/12827450876/job/35769769947?pr=754 ) ; not sure what the best course of action is, reinstate a --trace-output flag or fix the invocation in the scripts using -- ?

[AltGr]

yes that's the caveat I pointed to before: we should fix the scripts, either by expliciting -t - or by just moving the flag so that it doesn't appear right before the catala file name.
When we automatically handle the .json file extension, we should also add a warning pointing the correct course of action if we detect that the supposed trace file ends with .catala_?? (or even an error, to be totally certain that a catala file won't get overwritten!)

[rprimet]

Now that the catala-example CI has a candidate fix ( CatalaLang/catala-examples#19 ), the main issue remaining in the CI is that some messages have a different format, due to a few issues of Message.log being replaced by straight Format.fprintf in the code of print_log (see for instance git diff HEAD..master compiler/shared_ast/interpreter.ml) ; as a result, there are a few line break inconsistencies and missing [LOG] prefixes. @AltGr any ideas on what may be the best fix, parametrize Message so that the ppf instance can be user-supplied or something else?

[AltGr]

I wrote a workaround so that the format, when printing to the console, is the same as before (when in a file the [LOG] prefixes are removed, as they should be)

[AltGr]

@rprimet basically the last bug was that we were duplicating the formatter generated on top of stdout, which led to erratic behaviours when both versions were being used; should be all OK now.

[rprimet]

Thanks @AltGr !