[WIP] Print actual results with --print

[schoettl]

Closes #21

CLI proposal

Print test file:
     --print[=FORMAT]   Print test files in specified format (default: v3).

[deferred:]
     --actual[=MODE]    Combined with --print, print test files with actual
                        results (stdout, stderr, exit status). Mode 'all'
                        prints all actual results (default). Mode 'update'
                        prints actual results only for non-matching results,
                        i.e. regular expressions in tests are retained.

TODO

• [x] Refactor printShellTest
• [x] Fix bug: type of expected result is not considered => regex are also put in the next line instead of directly after >=/>/>2
• [x] Support formats other than v3?
• [x] Fix parser to simply parse comments and blank lines without discarding the leading comment marker # or *.
• [x] Don't discard trailing comments after the last test
• [x] Create shelltests for --print
• [x] Print shared input as it was defined originally
• [x] shelltests for --print --actual
• [ ] Don't discard comments at begin of file (before first test)

[schoettl]

I need help the parsing of a comment line:

The parsed comment line discards the comment character (# or *).

I think it would be better to store the full comment line for this reasons:

• When printing the shelltest file, I cannot restore the comment line without information about the comment character
• The code for printing would be simpler (no case matching, just putStrLn comment)
• The parsed commentline isn't used anywhere else anyway.

But I couldn't figure out how to change that line so that commentline stores the full line.

[simonmichael]

Here's a rewrite of commentline that should work, though it could be made much more compact:

commentline = try (do
    prefix <- many1 (char '*') <|> (whitespace >> many1 (char '#'))
    rest <- lineoreof
    return $ prefix ++ rest
  ) <?> "commentline"

[schoettl]

I know, this are a lot of changes already. I implemented your suggestion to parse commentline.

I also added a field trailingComments that are normally empty except for the last test. The alternative would be an extra data type only for trailing comments.

As far as I see, the biggest remaining problem is the "shared input". I don't even understand how the syntax works (haven't analyzed the parser yet). Currently, --print prints "shared input" as specific input for every test.

[simonmichael]

@schoettl I had a go at removing the actual output printing for now, but better to let you do it if possible. Would you mind removing just that part, and combining the rest into a single commit rebased on latest master ?

[simonmichael]

Oh shared input. It just means that after an input you can write more than one tests, and they'll all use it. Sorry for the hassle, but we will need to preserve this in the output.

[schoettl]

@schoettl I had a go at removing the actual output printing for now, but better to let you do it if possible. Would you mind removing just that part, and combining the rest into a single commit rebased on latest master ?

Alright, I'll try to separate it.

[schoettl]

Oh shared input. It just means that after an input you can write more than one tests, and they'll all use it. Sorry for the hassle, but we will need to preserve this in the output.

Yeah... that will mean more changes in the parser and probably two constructors for ShellTest.

Like data ShellTest = ShellTest { ... } | SharedInput { ... }

Do you agree?

[simonmichael]

two constructors for ShellTest

I don't have insight on that right now.. if you think it's the best option, let's give it a try.

[simonmichael]

I see what you mean, currently we don't model shared inputs in the type. If you want, that can be postponed for a future PR.

[schoettl]

I see what you mean, currently we don't model shared inputs in the type. If you want, that can be postponed for a future PR.

Yup, I think that's a good idea.

[schoettl]

I re-implemented the --actual options based on the --print already merged PR.

This feature requires --print option and prints an error if used without --print.

I'm open to changes about the option naming. Alternatives I considered:

• --actual[=spec] because it prints actual results
• --results[=spec] although it's very generic (test results, test reports, results of commands, ...)
• --print-with[=spec] – the word print is redundant, when --print is already given. Using it instead of --print[=format] is also not a nice CLI because for me print is the main action.

FYI, the options around --print are in their own option group, see -h. So I think it doesn't matter if the new option doesn't contain the word "print".

I would add tests and doc, if this PR is a merge candidate.

[schoettl]

The concept of this addition to --print is this:

Either non-matching-result matching-result

where result can be stdin/stdout/exit status.

From that and depending on the --actual[=spec], the Matcher is computed so that

• it's the original result from the input shelltest file
• it's the original Matcher where the test results match and the actual result where it doesn't – given spec = "update"
• it's always the actual – given spec = "all"

[simonmichael]

Blocker: #26

[schoettl]

I will rebase when PR for #26 is merged.

[schoettl]

The comments at begin of file are still not parsed. Thats because formatXtestgroup does skipMany whitespaceorcommentline before it parses many $ try (formatXtest i).

I don't quite understand this part of code. @simonmichael maybe you can look into it if you have time.

[simonmichael]

This is conflicting now - I had a look at resolving it but it's a bit much for me to tackle just at the moment, sorry.