reduce/resuce precedence

[mrgleba]

Path to address:
#40

Some discussion in my previous PR:
#47

Closes #40

[mrgleba]

I'd like to add that this is a potentially breaking change (there could be parsers generated differently after the patch). I would appreciate someone's second opinion.

[kkm000]

Thanks for the fix, LGTM. For its being potentially breaking, I believe it should be ok--the users have been warned about the conflict, and then a reduce/reduce conflict is always a thing to take seriously. Also, the next version is going to be 7.0, indicating some breaking would be expected.

It would be awesome if you added a test with the grammar from #40. Can you find time to add that?

[dsyme]

Please, please add a command line option to optionally preserve the exact old behaviour.

Maintainers: please don't accept this PR until this command line flag is added :)

This is to ensure that any downstream users get complete control over whether they adopt this change or not, "just in case".

Thank you!

[dsyme]

@enricosada Please note that the F# compiler repo would need to apply this flag, at least in a first instance until a thorough impact analysis has been done

[mrgleba]

I'll try to extend this PR to address #39 too.
I'll create a command line option that would allow to persist current behaviour in both cases (reduce/reduce and shift/reduce on nonassoc).

[dsyme]

Thanks!

[mrgleba]

I've added a new PR with both fixes and a command line option to preserve previous behaviour.
Please do not merge yet. I'd still like to investigate a thing or two here.

[kkm000]

Thanks!

Normally, you'd put "WIP" into the title of the PR to indicate it is work in progress and not to be merged.

[mrgleba]

I've looked into the way fsyacc tries to do error recovery.
In most cases this would propagate error up the state stack.
In other cases this would accept an invalid input
The before mentioned example of nonassoc operator is an example of the latter.
Another case is the possibility to accept (fnish parsing) before reaching EOF.

I'd vote to remove his behaviour. If someone needs it, it's still available with the new --oldprec option.
Alternatively I can and another option to control this behaviour.
I'd preferr the former, but I'll gladly hear your opinion.

I'll try to find time to add some tests to this commit and get back to you to merge the whole PR.

[mrgleba]

I've finally found some time to get back to these changes.
The new commit has 2 command line args:
--newprec - to ebable new precedence calculation mode
--no-recovery - to disable reduction as a mode of receoverying from invalid input

With neither of the above fsyacc would work as previously (the overal change is not-breaking - you have to opt-in to get the new functionality).

[dsyme]

Closing this old PR