Add a --output/-o option to specify mdx test output

[NathanReb]

Fixes #193

This PR adds a --output/-o option to set the output for the ocaml-mdx test command.

It used to always write to <input_file>.corrected so the default behaviour is preserved, i.e. if you do not pass the --output option, it still writes to that file.
If passed --output -, the command will write the corrected file to stdout instead of writing it to a file. Setting stdout as the output implies --force-output. This seems like the best behaviour if you want to pipe it into some other command or redirect it to a file for later diffing.

I deleted the code of Misc.run_expect_test (which isn't part of the pulbic API) since it was only used in Mdx.run. Instead I created a more generic function Mdx.run_str which returns a pair composed of the test result (whether the corrected version is equal to the input or not) and the corrected content.

It is used by two new functions Mdx.run_to_stdout and Mdx.run_to_file. The former just writes the corrected content to stdout no matter what, the second preserves the behaviour of Misc.run except that it takes an outfile argument.

I wasn't too sure about what needed to be done with the "clean up" behaviour which takes care of deleting the corrected file if it exists and the test succeeded so I kept it for any output file, whether it's the default are an explicitly specified one. I don't think it's necessary to do that when used through dune as it doesn't use the file presence alone to determine which files need to be promoted but that's unrelated to this PR.

Finally, to preserve the public API, I exposed these two new functions and re-implemented Mdx.run on top of Mdx.run_to_file so that its behaviour and signature remain the same.

I also added tests to ensure that this option works properly.

Let me know what you think!

[Julow]

Should --output implies --force-output in every cases ?
If I specify the output file on the CLI, I don't expect it to be not create or even removed.

[samoht]

Can you use the compat layer to keep support for 4.02? Thanks

# File "bin/cli.ml", line 113, characters 17-19:
# Error: Unbound constructor Ok

[NathanReb]

I think force-output should always be true, that would spare some effort on our hand and offer a more consistent behaviour but let's keep that for a separate PR!

[NathanReb]

Can you use the compat layer to keep support for 4.02? Thanks

Done!

[Julow]

I think force-output should always be true, that would spare some effort on our hand and offer a more consistent behaviour but let's keep that for a separate PR!

I meant setting force-output when --output is set. Changing the default behavior requires an other PR but this is the behavior of --output that I'm talking about. Let's not implement legacy code.

[NathanReb]

I understood what you meant, my point was that we can just tackle one thing at a time. Having --force-output implied when output is specified means having completely different behaviours for --output and the default which currently is just --output <input>.corrected.

I'd rather drop force-output completely in a future release. Let's keep the breaking changes to a minimum for the next one.

[Julow]

I don't agree. The default feels like --output <input>.corrected but it doesn't have to be !
My point was that this has nothing to do with the default behavior and it should be discussed and implemented correctly the first time.
This behavior (not writing a file, or worse, removing it when, as a user, I wrote it on the command-line) is not right for me.
You already implemented a special case for --output -.

[NathanReb]

Yeah it's a good point, I just fixed it!

[Julow]

Looks good !