[PATCH]: ResourceManager bug fix

[Paulo Gaspar <pa...@krankikom.de>]

In the ResourceManager class, 
  resource.process()
was always (twice) called before the function call
  resourceLoader.getLastModified( resource )
that gets the resource timestamp.


The problem is that if (with loads of bad luck) the 
resource is updated between these too calls, the old 
resource version data would be associated with the 
newer version timestamp.

The consequence is that the subsequent timestamp 
checks on the resource would give it has not changed.


These patch is not ideal but it is what is possible
with the current architecture (*). It simply consists 
in an order change - getting the timestamp before 
loading the resource.

Its worse case is that a new version of the resource,
changed (again) between the 2 calls will be loaded 
twice - because of the new version data being 
associated with the old timestamp. 

However, this just results in redundant work on the
very-bad-Murphy-at-work case instead of deficient
functionality (e.g.: an updated template not being 
reloaded).


As mentioned before, be aware that text newline 
sequences might be wrong for Linux (I'm not sure). 
And I am sure it will need some editing to work. 

Can some one fix it and apply it?


Thanks and have fun,
Paulo Gaspar

(*) - Anyway, I think it is impossible to read the
      contents of a file and its timestamp in a 
      single step. So, the most frequent case of 
      the file would always have this limitation,
      unlike database stored resources.

RE: [PATCH]: ResourceManager bug fix

[Paulo Gaspar <pa...@krankikom.de>]

Thanks for applying those patches Geir.

On the Linux remark: I want to try it for some years. I already 
worked a (very little) bit with Unix systems and - more 
important - with people that knew how to use them. So, I am 
aware of the advantages... but I am also too much aware of the 
time it will take to learn it.

The last advantage I got aware of, was on how much more stable 
Linux is. Here, at my job, people consider normal never having 
to reboot a server that - to make things harder - generates a 
load of dynamic content. Last year, our most loaded server was 
rebooted _once_... intentionally (a kernel upgrade or something 
like that).

Now, any Windows NT experienced guy reboots a loaded NT4 server 
weekly or every other week just to be sure that the software 
running on it doesn't crash. (There is a saying that crashing a 
piece of software on Windows is just a matter of time - which,
fortunately, is getting longer but is still measurable.)
 

Anyway, Open Source stuff and Cygnus are giving me a start. 
Back to good old command line tools... but for something better 
than DOS.


Have fun,
Paulo

> -----Original Message-----
> From: gmj@mta6.srv.hcvlny.cv.net [mailto:gmj@mta6.srv.hcvlny.cv.net]On
> Behalf Of Geir Magnusson Jr.
> 
> Paulo Gaspar wrote:
>
> ...
>
> > As mentioned before, be aware that text newline
> > sequences might be wrong for Linux (I'm not sure).
> > And I am sure it will need some editing to work.
> 
> You should give Linux a whirl.  You can't beat the pricepoint...
>
>...

Re: [PATCH]: ResourceManager bug fix

["Geir Magnusson Jr." <ge...@optonline.net>]

Paulo Gaspar wrote:
> 
> In the ResourceManager class,
>   resource.process()
> was always (twice) called before the function call
>   resourceLoader.getLastModified( resource )
> that gets the resource timestamp.
> 
> The problem is that if (with loads of bad luck) the
> resource is updated between these too calls, the old
> resource version data would be associated with the
> newer version timestamp.

Yes, that would be loads of bad luck.  But its true - its possible.

> 
> The consequence is that the subsequent timestamp
> checks on the resource would give it has not changed.
> 
> These patch is not ideal but it is what is possible
> with the current architecture (*). It simply consists
> in an order change - getting the timestamp before
> loading the resource.
> 
> Its worse case is that a new version of the resource,
> changed (again) between the 2 calls will be loaded
> twice - because of the new version data being
> associated with the old timestamp.

Right - and that is acceptable as it beats the other issue, of not
having the most up-to-date resource data.
 
> However, this just results in redundant work on the
> very-bad-Murphy-at-work case instead of deficient
> functionality (e.g.: an updated template not being
> reloaded).
> 
> As mentioned before, be aware that text newline
> sequences might be wrong for Linux (I'm not sure).
> And I am sure it will need some editing to work.

You should give Linux a whirl.  You can't beat the pricepoint...
 
> Can some one fix it and apply it?

Done.
 
> Thanks and have fun,

Always do.

geir

-- 
Geir Magnusson Jr.                               geirm@optonline.net
Developing for the web?  See http://jakarta.apache.org/velocity/