cruisecontrol

In the current design of writing build status to a file on disk, there 
is an intermittent bug that only happens on Windows. Windows doesn't 
like you writing to a file at the same time someone else is reading the 
file. (This is not a problem on Unix file systems).
The result is errors trying to update the currentBuildStatusFile if you 
happen to be doing so while a web app is reading the same 
currentBuildStatusFile.

The issue currently occurs rarely, but as I thought about implementing 
more features of the Progress logic (as in filtering of progress 
messages similar to how XmlProgressLogger does it), I realized we'd be 
likely to hit the problem more and more often.

That said, what do people think about:
1. Implementing some sort of locking on the currentBuildStatusFile (not 
sure how that would work, and even if it could work, given that any 
unknown process could be reading the file at any time).

2. Implementing some other approach to currentBuildStatusFile that is 
not file based (and thus avoids such concurrency issues). The major down 
side of this is that all the currently simple reporting options get a 
bit more complex than just having to read in the currentBuildStatusFile 
  text file. Any ideas about what this would look like?

3. Some other approach all together?

Maybe finding some way to minimize the number of times/length we have to 
write to currentBuildStatusFile would help (although that just hides the 
problem, but doesn't really solve it).

Thoughts?

Dan Rollo

Date: Fri, 18 Jul 2008 11:41:53 +0200
Subject: [Cruisecontrol-devel] nested ant call through java: how to
	preserve	progress?
To: cruisecontrol-devel@...

Hello,

I post this question to the developer list because I think it is rather 
a development issue.

Following scenario: we build our projects through eclipse PDE-build. 
This works so far, I even could establish a working CC 2.7.2 with 
progress on the main status page: providing the AntProgressLogger was a 
bit tricky, though.

The problem now is that our build file (the one I call through CC) is 
calling another build file but through a java call (basically an eclipse 
call with ant-syntax). From that point, I see the progress messages like 
"ccAntProgress -- ..." at the console, but the progress at the web page 
remains at the last called target of the origin build file (which is 
quite generic). I noticed during integrating the AntProgressLogger that 
the CC process ant the launched ant process are connected through a pipe 
(STDIN of CC is connected to STDOUT of ant). I unfortunatly don't have 
the faintest idea how to pipe the output of the launched sub-process to 
the parent process. Is there a point I'm missing?

Yours,

Frank


Hi Frank,

I don't think you are missing anything, and reading STDOUT is exactly 
how the progress messages work for the AntBuilder.

Maybe you could add your own Ant Listener/Logger to the "another build 
through java call" that would read and forward the STDOUT of the 
"another build" to the STDOUT of the "build called through CC"? I'm not 
sure how exactly to set that up - more of an Ant trick than a CC thing - 
but as long as the output makes it to the "build called through CC" it 
should work.

You might want to extend the AntProgressLogger in some way (or aleast 
reference AntProgressLogger.MSG_PREFIX_ANT_PROGRESS) to ensure the 
format of the STDOUT messages (prefixed with ""ccAntProgress -- ") is 
correct. See AntProgressLogger for details.

Please let us know how this goes. I'm curious about chaining STDOUT like 
this, as it seems there may be other uses for such a trick... We may 
even want to refactor the AntProgressLogger stuff to make it easier to 
do this.

Dan

Oh dear god, please let's not go and make everything that could be
generics generics for the sake of being generics.

Jtf

On Sat, Jul 19, 2008 at 12:49 AM,  <PaulJulius@...> wrote:
...
>
> Log Message:
> -----------
> Now that we are using Java 5, we can use generics!
...

-- 
http://blog.jeffreyfredrick.com
http://www.ohloh.net/accounts/JeffreyFredrick