[Webkit-unassigned] [Bug 104858] "Running 1 DumpRenderTree over X shards" is not a helpful output

[bugzilla-daemon at webkit.org]

https://bugs.webkit.org/show_bug.cgi?id=104858





--- Comment #8 from Dirk Pranke <dpranke at chromium.org>  2012-12-13 09:59:12 PST ---
(In reply to comment #6)
> I agree that sometimes jargon can be good for its precision, but you have to consider the possibility that a particular term may confuse more than it enlightens.

Of course. That's why I said I wasn't necessarily making that argument.

> At least four WebKit contributors said on IRC that they actually found this confusing and misinterpreted the message. 
> 

At least as many people on IRC said that they thought this was a correct and accurate term. And it may be that now that the others have had this explained, they are enlightened. I'm pretty sure that's how learning new vocabulary works.

> (Also, it's kind of crazy that NRWT has so many distinct ways to group the tests that it's hard to find a free term! I am curious what "part" and "chunk" refer to and why they different from "shard".)

>From new-run-webkit-tests --help:

    --run-chunk=RUN_CHUNK
                        Run a specified chunk (n:l), the nth of len l, of the
                        layout tests
    --run-part=RUN_PART
                        Run a specified part (n:m), the nth of m parts, of the
                        layout tests
    --batch-size=BATCH_SIZE
                        Run a the tests in batches (n), after every n tests,
                        DumpRenderTree is relaunched.

"chunk" and "part" control which tests are run; sharding controls how they are run and is pretty much an internal detail that the user doesn't normally need to be aware of.

(In reply to comment #7)
> If a shard to DRT is almost always a directory, why not say: "running DRT in parallel over X directories" or something similar.

Because it's not always a directory. Even in the default case, we carve all of the http tests into a few shards (1/4 the number of DRTs by default), which means that multiple directories are usually in a single shard here. 

Finally, I will note that the code refers to "shards" in multiple places. Any change should update the code as well as the text string in the message.

I suggest that we either leave the wording alone, or make the sharding a debug message so that devs don't normally see it. I'm fine with either option. Learning a new use of a word is not a bad thing. I am also about to write more documentation about how NRWT works, and will need to refer to this concept as part of that (so it would be good to end this debate sooner rather than later). Perhaps having the documentation will partially address your concerns over confusing users not familiar with this usage.

If you are dead set on wanting to change this, of all of the alternatives that have been suggested, "batch" is probably the second-best word to describe things after "shard". I don't think the --batch-size flag is used very often, and we could and should probably rename it to something more descriptive like --restart-drt-after-every-n-tests or something.

I do not like any of those other suggestions that we use generic collective words like "part", "chunk", "group", or "section". There are already too many generic collections in NRWT and I have been working hard to get rid of them in order to make things clearer. Using one of these words would be a step backwards.

-- 
Configure bugmail: https://bugs.webkit.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.