Escape more characters in JSON output. Fixes #517

[stylpe]

517

Disclaimer: This was implemented entirely using GitHub Codespaces (without sufficient RAM to build the whole project) on my mobile phone during toilet breaks with zero experience in Haskell. Please be gentle!

[stylpe]

Ah, small mistake in the commit message, I meant to note that it only addresses the first point about the backslashes.

Also, I haven't checked if there's anywhere else where escaping is performed, which could lead to double escaping. Trusting the unit tests will catch that. (Maybe I should add my own unit test even)

[ozgurakgun]

Can you see the CI output @stylpe ? After clicking a few times, I get to this page: https://dev.azure.com/conjure-testing/conjure/_build/results?buildId=442&view=logs&jobId=b227660a-cc34-513e-56b6-a239a911c194&j=b227660a-cc34-513e-56b6-a239a911c194&t=778185b6-24a8-5c4e-9ea7-ecd373f7924f

Which gives a compilation error:

/home/vsts/work/1/s/src/Conjure/Language/Pretty.hs:124:13: error:
    • Variable not in scope: singleton :: Char -> Text
    • Perhaps you meant one of these:
        ‘T.singleton’ (imported from Data.Text),
        ‘V.singleton’ (imported from Data.Vector),
        ‘M.singleton’ (imported from Data.HashMap.Strict)
    |
124 | jsonEsc c = singleton c
    |             ^^^^^^^^^

[stylpe]

Yep, already made one amendment based on the CI logs :)

The error message is particularly helpful this time, it correctly suggests T.singleton as a fix! Will commit and push in a moment.

[stylpe]

And look, the language service finished installing overnight so I can see it in the editor now!

[stylpe]

Ha, speaking of insufficient RAM, you might need to size up the builders:

ghc.exe: getMBlocks: VirtualAlloc MEM_COMMIT failed: The paging file is too small for this operation to complete.

[ozgurakgun]

The windows builder on Azure has been unreliable, yes. The linux/max ones should work fine I am hoping.

[stylpe]

Here's a minimal failing repro model, but I'm not 100% sure where to add it, is tests/pretty_print/issues/ ok?

$ The variant triggers refinement output that is vulnerable to the bug
find x : variant {i:int(1..3),b:bool}
$ The backslash at the end of the next line doesn't get eacaped, producing "... /\"
such that active(x,i) -> x[i]>2 /\
x[i]<5

$ Simpler cases that fail since `\ ` isn't a valid JSON escape sequence
find y:bool such that y /\ true
find z:bool such that and([z,true])

[ozgurakgun]

The parse_print directory is quite specialised for testing the roundtrip: parse/print/parse is the same as parse.

You can put it under tests/custom/issues/519 if you like? You'll have to put a run.sh file in there as well.

[stylpe]

I learned how to use git rebase -i, so I fixed up the branch a bit. The added test is a blind commit again, so relying on CI for feedback, otherwise I consider this ready!

[stylpe]

This time it should be OK!

[ozgurakgun]

Thanks for the contribution @stylpe!