[eclipse.org-architecture-council] [Bug 246840] New: [discussion] How to

[bugzilla-daemon@xxxxxxxxxxx]

https://bugs.eclipse.org/bugs/show_bug.cgi?id=246840  
Product/Component: Platform / Runtime
           Summary: [discussion] How to use Jobs and ISchedulingRule
                    properly?
           Product: Platform
           Version: 3.4
          Platform: All
        OS/Version: All
            Status: NEW
          Severity: normal
          Priority: P3
         Component: Runtime
        AssignedTo: platform-runtime-inbox@xxxxxxxxxxx
        ReportedBy: martin.oberhuber@xxxxxxxxxxxxx
                CC: john_arthorne@xxxxxxxxxx, markus.schorn@xxxxxxxxxxxxx,
                    eclipse.org-architecture-council@xxxxxxxxxxx
        Depends on: 246839


Javadocs for ISchedulingRule.isConflicting() vs. ISchedulingRule.contains() are
there, but missing on the point how it is supposed to be used. How to write a
contains() rule properly? How to use it with RuleFactory and MultiRule?

This question stems from bug 240888, which tries to save Project Preferences
properly in a Job. How to declare a SchedulingRule that allows this run
concurrently with other projects? We need to run a job with a project-rule.
Additionally in the job some project-preferences will be flushed. The flushing
will potentially lock some resources, outside of the project!!.

  - Problem 1: SchedulingRules need to be created in advance, and cannot be
    changed/added/extended later on. So, the creator of the Job has to know 
    in advance, that the flushing of the preferences will lock resources.
    I need to look at its implementation to figure out the maximum of
    resources that will be locked. This is against information hiding (and may
    break in the future, when implementation is changed).

  - Problem 2: The rules are determined via IWorkspace.getRuleFactory(). It
    turns out that the rules I obtain from there can differ from call to call.
    This makes the situation worse in that I can run into an 
    IllegalArgumentException, even if I obtained the rules for the job in the
    same way as it is done in the methods I am using inside of the job. How to
    use it properly? 

Bug 240888 is currently marked fixed, it was solved by taking a Workspace rule.
But is this the best we can do? 

The general problem seems to be that the creator of the Job needs to know in
advance what locks are required, even if his job calls out to other (unknown)
items. 

  * Is this restriction really required in all cases? Or could some
    "specially marked" SchedulingRules allow extension later on in 
    the Job, for rules that are safe because they cannot be taken
    "in the wrong order"?

  * Should it be part of API Docs what SchedulingRules some API Call 
    requires? This likely wouldn't work with synchronous callbacks.

  * Is there some different usage pattern to avoid the problem? 
    How to use RuleFactory and MultiRule properly?

There also seem to be some semantic problems with ISchedulingRule#contains(),
which I filed separately (see bug 246839).


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