I fixed the regexp for Go 1.5 as explained on https://golang.org/pkg/runtime/

Currently, it is: gc # @#s #%: #+...+# ms clock, #+...+# ms cpu, #->#-># MB, # MB goal, # P

I ended up also adding tests and fixing a few things:

• It was parsing Stderr and printing on Stdout (via Println). It now always write to Stderr.
• I added a flag to be able to set a fixed port.

When the parser signals that it is done, the main loop wants to exit, but the break for the for loop was only breaking the select case and not the loop.

Before this patch, if the input was closed (for example when we pipe into gcvis and the process on the left was done) gcvis would always go into the done case, just break the case and then continue to the next iteration of the loop, which would again choose the done case and continue like this forever, eating up a single CPU core (effectively busy looping).

Fix this by replacing the break with os.Exit(0).

I wanted to get insights in the GC timings so I extended gcvis to also parse out the timing measurements and show them. note: i just modified the regex for go 1.5, but only tested it with 1.6, not sure if these versions use an identical format? screenshots (note: program instrumented had very few scavenger runs) BEFORE: AFTER:

So I was looking at gcvis and noticed that it didn't work with my app... So I dug into it and found out that when I use goroutines it would show nothing.

I ended up noticing a previous PR https://github.com/davecheney/gcvis/pull/9 on your repository but which was causing previous output to not be parsed anymore. At this point, I decided to just add tests on the project (at least on the Parser part).

I couldn't help but also refactor the whole main.go file to make a bit more sense of it and get to write only tests on the Parser part.

A bit of cleanup is still necessary IMO, but I'll get to that later. Would you be okay to review/merge this PR?

The javascript is currently broken.

This one is on me :disappointed:. When I refactored I was still running the old gcvis executable in my GOPATH. Turns out goimports added this dependency thinking it was a text/template when in fact we used html/template before.

Sorry about that. I wish I could have a bit of time to add tests around the http server but that'll have to wait a bit.

This allows us to match on goroutines conditionally and still support go versions inferior to 1.4.

If you could review/merge this one, then I could rebase my other PR on top of master to fix the timeout issue everywhere.

Cheers.

Work in progress.

This PR introduces the possibility of running gcvis with the content of Stdin.

@davecheney if you have any advice on adding integration tests on the CLI, I'm all ears :smile:.

timeoutAfter would make other tests cases fail if they took more than the duration passed in parameter. Each test now creates its own timeout channel, avoiding shared state between tests.

@davecheney You told me to add the timeout in runParserWith but I feel that placing it in timeoutAfter makes more sense.

Issue: When using pipe call scheme (not a subcommand one), title of the page was not being set properly (it was rendered as gcvis -)

Here I suggest setting it to iface:port in such cases (eg. for localhost:6060 the page title would render as gcvis - localhost:6060)

I wanted to see accumulated totals. This isn't ideal from a memory POV, but it got me the data I needed.

Also refactored the JS a bit, seems to work.

To enable someone to use this tool to observe event streams over sliding windows of sample recency as well as prevent unbound growth of observations, the list of observations has been wrapped in a ring buffer of flag-specifiable size. This has been a tremendous help in visualizing events from particularly garbage collector-active servers.

This is a PR to fix #1, there are some limitations outlined bellow and I'll see if I can fix the transparent image later on.

Limitations: Currently FF saves with incorrect image extension and the background of the image is transparent (will fix later).