stack bench truncates final output on a slow terminal

[simonmichael]

If a benchmark executable prints much output, in a slow terminal (such as an emacs shell buffer) "stack bench" truncates it.

[snoyberg]

Do you have a repo for this? stack bench just calls runghc Setup.hs bench under the surface, which I thought sent all of the output to a log file.

[simonmichael]

Hey Michael, https://github.com/simonmichael/hledger/ master should show the problem now if you uncomment hledger/bench/bench.hs line 30 or 31 and run stack bench hledger in an emacs shell buffer (M-x shell in emacs). You can see the full non-truncated output eg with:

(cd hledger; stack exec -- ghc -ibench bench/bench && stack exec -- bench/bench --simplebench)`

[snoyberg]

I didn't realize that Cabal's bench output went to the terminal, interesting.

[snoyberg]

Since we're just reusing Cabal's run code here, can you tell me if the same problem appears when you would run cabal bench?

[simonmichael]

I don't see the problem when running cabal bench myself.

[snoyberg]

@chrisdone Do you have any thoughts on this?

[snoyberg]

I just merged a PR that changes the buffering of stdout, I'm not sure if that will have any impact. For the record, I've been unable to repro so far.

[simonmichael]

Today's stack (revision 2cac22d) has the same problem here, FWIW.

I think maybe Stack.Build.Execute.printBuildOutput should hold on to the printing thread's id, so that the main thread can wait for it to finish before exiting ?

[simonmichael]

PS awkward that you can't reproduce. To be more specific, I see this in a shell buffer in GNU Emacs 25.0.50.1 running in the normal GUI mode on OSX Yosemite on a 2013 macbook air.

[radix]

Emacs for OS X has really slow buffer drawing compared to Linux, so it may be difficult to repro on Linux vs Mac.

[snoyberg]

Can you think of any other way to reproduce this problem? As it stands right now, I don't think anyone else can attempt to fix this.