How polished do tools need to be?

Jonathan Gibbons jonathan.gibbons at oracle.com
Sun Jun 23 10:01:45 PDT 2013


On 06/22/2013 09:43 AM, Alan Bateman wrote:
>
> friday-stats is a great idea!!
>
> So I'm curious how "polished" tools need to be? The initial tools that 
> Jon pushed are really nice, ready to use with any CI system.
>
> I have a several scripts that that I use to get interesting data and 
> I'm wondering whether to clean them up. For example, I have a script 
> that runs over the jtreg work directory to get print out the 50 
> slowest tests (I'm pretty sure Jon has a script for that one). Another 
> one is a script that summarizes the ProblemList.txt to give the count 
> of excluded tests by platform. Some of these might be too trivial but 
> I find them useful.
>
> -Alan
>
>

Well, that is an interesting question for discussion...

At some level, I think the most important issue is to provide a home for 
utilities that developers have written which they have find helpful to 
use on an ongoing basis to get information about the OpenJDK source code 
and builds.  (So a script to achieve a one-off task does not count -- at 
least, not until you adapt it to use it a few more times.)

But, there clearly have to be some guidelines. How about ...

* Should be usable with (just) the general toolset used by OpenJDK 
developers.
* Should not use native code, to keep build issues at bay.
* Should be parameterized via command line options
* Should be reasonably well documented

My personal preference is to provide such tools as Java utilities, since 
it is easier to achieve a higher quality in Java, compared to a shell 
script. But I also accept that many utilities start off as shell 
scripts, and then "just grow".  All of the utilities I've posted so far 
started off as scripts. The question is, at what point do they 
metamorphose from their larval state to a magnificent butterfly?

I think potential integration into a CI system is a good goal to aspire 
too. It is one thing for an individual to run tools occasionally, if and 
when they remember to do so. I think it is more interesting to use CI to 
run these tools on a regular basis and then provide a way to aggregate 
the results (or links to results) onto a single page per forest.  (And, 
as a secondary point, tools are less likely to suffer bitrot if they are 
run regularly by a CI system.) I also think that one of the reasons for 
this project is to provide exposure to info which has been implicitly 
available but hidden for far too long.




More information about the friday-stats-dev mailing list