Add line wrap disablement to ConsoleReporter

[edmundnoble]

I don't propose changing cabal.project, this is just for a demo until the PR to ansi-terminal is merged (UnkindPartition/ansi-terminal#171)

I saw that our test suite leaves garbage in the terminal output despite using --hide-successes. This turns out to be worse the narrower my terminal is. The more general problem is that ConsoleReporter leaves garbage in the terminal when the lines it prints are wrapped, because it doesn't know to clear multiple lines, just the most recent one. This PR fixes that by disabling line wrapping while printing that line.

This PR makes ConsoleReporter's ANSI tricks clean up after themselves properly in the presence of a narrow terminal, a long test name, or a long progress message, including those printed by testCaseWithSteps from tasty-hunit. It also prevents test group names from wrapping, because that has the same cause.

[Bodigrim]

That's a nice idea! Let's wait for UnkindPartition/ansi-terminal#171 to be released, then I'll prepare a more in-depth review.

[Bodigrim]

Shall we use bracket?

[edmundnoble]

I don't know really what the best practice is here. Printing is quick so I don't see it being interrupted by an async exception. Printing could result in an exception caused by stdout becoming unavailable, but in that case we can't print the enableLineWrap code anyway, so there is no recovery.

It does seem natural to use bracket though. It's up to you.

[edmundnoble]

Did it for now.

[Bodigrim]

Please update the Cabal file with ansi-terminal >= 1.1.2.

[Bodigrim]

Or, even better, let's guard usage of disableLineWrap / enableLineWrap by #if MIN_VERSION_ansi_terminal(1,1,2) and use no-op instead otherwise.

[edmundnoble]

@Bodigrim ansi-terminal 1.1.2 has been released and I've updated this PR to have it as a minimum bound. 🥳

Edit: I didn't see your above comments, github issue.

[edmundnoble]

The latest update should address all feedback thus far.

[Bodigrim]

We certainly benefit from no line wrapping when executing tests with --hide-successes. But do we benefit from disabling line wrapping in the absence of --hide-successes? Maybe it should be used only when --hide-successes is enabled?..

[edmundnoble]

We certainly benefit from no line wrapping when executing tests with --hide-successes. But do we benefit from disabling line wrapping in the absence of --hide-successes? Maybe it should be used only when --hide-successes is enabled?..

I argue that we do benefit from disabling line wrapping without --hide-successes: progress indicators get cleared even without --hide-successes, so this bug remains, especially using testCaseWithSteps. Also when printing test group names, we keep indenting further to the right, and on a narrow terminal, this may cause wrapping for longer test and group names.

[Bodigrim]

I argue that we do benefit from disabling line wrapping without --hide-successes: progress indicators get cleared even without --hide-successes, so this bug remains, especially using testCaseWithSteps.

That's a good argument.

My main concern is that without line wrapping resultShortDescription will be invisible on a narrow terminal. Which is especially undesirable if the result is FAIL. I think we should switch line wrapping on before printing resultShortDescription always or at least in case it is FAIL.

[edmundnoble]

I argue that we do benefit from disabling line wrapping without --hide-successes: progress indicators get cleared even without --hide-successes, so this bug remains, especially using testCaseWithSteps.

That's a good argument.

My main concern is that without line wrapping resultShortDescription will be invisible on a narrow terminal. Which is especially undesirable if the result is FAIL. I think we should switch line wrapping on before printing resultShortDescription always or at least in case it is FAIL.

That's an excellent point. I undid that part of the change, now the result is not wrapped.

[Bodigrim]

Thanks, that's awesome!