We test for whether the sentinel is present, but we do two separate
filtration passes (first in dev/core for About.{java,properties}, and the
second in distro-source for the about.{txt,html}). If we had only one file,
the first pass would touch it, and the second would see it as "available,"
might see the contents as up-to-date (e.g. if the source about.{txt,html}
were untouched, destination output newer than that, but still containing the
"old" SVN info)... net effect, only the first filtration would reliably
happen without a clean start.
On Fri, Sep 5, 2008 at 7:14 PM, Scott Blum <[EMAIL PROTECTED]> wrote:
> I haven't fully reviewed the changes, particularly to /common.ant.xml; but
> I am curious why we need two different sentinel files rather than one file
> with a composite name?
>
>
> On Fri, Sep 5, 2008 at 4:41 PM, Freeland Abbott <
> [EMAIL PROTECTED]> wrote:
>
>> No, I didn't test the timing impact, and certainly not over several
>> cycles. Yeah, I didn't realize it'd have the trickle-down... I thought user
>> was an independent compile, and didn't think about the samples at all.
>>
>> Sigh.
>>
>> Re-done with sentinels restored and expanded for svnrev. Note that the
>> svnrev value might contain characters illegal in Windows filenames, so
>> there's an extra hop for that... also, we need to break the svninfo-getting
>> bits out from gwt.revfilter, because we need that info to know what
>> sentinels to look for to decide whether or not to filter. Finally, the
>> about.{txt,html} were never sentinel-protected, which I think might make
>> them wrong (in that an old version could be generated, a svn up performed,
>> and the new one not generated) and certainly adds some time, so I
>> sentinel'ed them along the same pattern.
>>
>>
>>
>>
>>
>>
>> On Thu, Sep 4, 2008 at 1:25 PM, Scott Blum <[EMAIL PROTECTED]> wrote:
>>
>>> Freeland, did you actually try this out? You're killing the desk-rebuild
>>> experience here, because when you force gwt-dev to rebuild every time, that
>>> trickles down into a rebuild of user, samples, the whole nine yards.
>>>
>>> Adding gwt.svnrev into the name of the sentinel file would be a much
>>> better alternative.
>>>
>>> On Thu, Sep 4, 2008 at 3:19 AM, Freeland Abbott <
>>> [EMAIL PROTECTED]> wrote:
>>>
>>>> That's as good a cause for a -1 as any, I suppose... attached no longer
>>>> uses non-static "this" in static context, and also untangles (by removal)
>>>> the creation of a sentinel file to gate the filtering. I believe the
>>>> sentinel exists so that a change in gwt.version will cause the extant file
>>>> to be considered out-of-date regardless of timestamp, but we'd now need
>>>> another one for change in gwt.svnrev, and it seems the original test (much
>>>> less the expanded one!) is probably more expensive than just refiltering
>>>> each time!
>>>>
>>>> On Wed, Sep 3, 2008 at 2:24 PM, Scott Blum <[EMAIL PROTECTED]> wrote:
>>>>
>>>>> 'Fraid I'm going to have to give this a -1 on account of not compiling.
>>>>> :)
>>>>>
>>>>> On Wed, Sep 3, 2008 at 12:23 PM, Freeland Abbott <
>>>>> [EMAIL PROTECTED]> wrote:
>>>>>
>>>>>> Rajeev noted that my initializer in About.java is an instance
>>>>>> initializer, not (as intended) a static one. Things appear to work
>>>>>> anyway
>>>>>> because they pull from about.txt and .html, but programmatic access
>>>>>> would be
>>>>>> uninitialized....
>>>>>>
>>>>>>
>>>>>
>>>>
>>>
>>
>
--~--~---------~--~----~------------~-------~--~----~
http://groups.google.com/group/Google-Web-Toolkit-Contributors
-~----------~----~----~----~------~----~------~--~---
