Hi Ed,
El 06/02/2012 20:05, Ed Willink escribió:
Hi Adolfo
At least we have got an agreement, the current
impact analyzer should be promoted to org.eclipse.ocl.ecore :)
I do not know what you mean, so I cannot be in agreement.
I don't see the Impact Analyzer as part of the
org.eclipse.ocl.ecore plugin; we must not add so much code to
existing functionality.
I don't see the Impact Analyzer as part of the
org.eclipse.ocl.ecore feature; we must not add so much code to
existing functionality.
I do see the Impact Analyzer as part of the Core build.
I see promotion as dependent on a satisfactory review, so I do not
yet agree that the current Impact Analyzer should be promoted.
I also see promotion as dependent on a decision on namespace
issues.
I should have said "promoted under org.eclipse.ocl.ecore" namespace,
that is, org.eclipse.ocl.ecore.impact*, which was originally your
proposition. However, I misread that you changed to
"org.eclipse.ocl.impact" so, as you said we are not in any agreement
after all ;P
I've always seen the OCL Tools components as those
UI-based components which rely on a OCL Core infrastructure, as
they were thought and we had been developed in the currently
terminated OCL Tools project. Whereas the OCL Core components
could sensibly work in stand-alone mode, the Tools components
could probably need some Eclipse UI platform to run. Remember
the namespace used to distinct bundles between the "MDT/UML"
project and the "MDT/UML Tools" one.
The MDT/UML2 split is easy; the UML project supports just models
in UML, CMOF and Ecore forms. Everything else is a tool.
MDT/OCL already has some tools as core; the parser, analyzer and
evaluator are 'tools'. The Impact Analyzer is a tool.
I see a Tool as a coherent module of additional functionality,
which may or may not have a UI. The Impact Analyzer just happens
to have no UI at present. I suspect that some profiling might
support interactive optimisation.
Being fussy, most Java classes could be a tool as well. I'm not
going to go into such a pedantic discussion. To me, as final Eclipse
user I prefer to consider a Tool something I can interact with, i.e.
ui-based stuff. To do a simil, JGit would be the Core, EGit (which
will probably have its core components) would be the Tool.
On the one hand, distinguishing Tools components
with a prefix does quite sense to me. Looking at a Juno
installation, excepting the "org.eclipse.e4.emf.xpath" bundle,
every bundle of the "new" e4 project are categorized in a "core"
/ "ui" fashion. On the other hand, projects like EMF have its UI
components categorized in the respective feature as you
suggests.
In my hypothetical 'proof' tool I tried to address this. UI/not-UI
is a different issue to Core/Tools.
Yes they are different, but that's what I want to map. Non-ui stuff
prefixed with core and packaged as Core components, Ui stuff
prefixed with ui, and packaged as UI components.
All
the xtext editors come as two plugins; a non-UI and a UI one. So
every Xtext editor splits across your Core/Tools boundary.
Indeed, as I originally suggested. On the other hand, I'm fearing
that the "core" (non-ui) part of the current editors (basically the
parsers and evaluators) may depend on Xtext infrastructure, hence,
there would not be any chance to include the new
languages/parsers/evaluators into our Core distribution. So the
mapping I mentioned above is not so clear.
I prefer the "core/tools" prefixes since it clearly
distinguish core from UI-based components, apart from the
benefits I mentioned in my previous post. Perhaps "ui" sounds
better for you. Perhaps we could not introduce the "core"
prefix, but at least use "tools/ui" one.
(I don't like '.ui' but anything else would now be worse.)
Note that EMF uses suffixes: ".edit", ".editor", ".impl" not
prefixes.
What is your definition of 'Core'? I suspect it is what is
packaged in the Core build. Packaging can change, tools change
much less, so I don't like naming tools according to today's
packaging policy. Whether EMF's ".edit", ".editor" are in
separate plugins is a packaging decision that can be changed
without renaming.
If you still have doubts what Core means to me/how we should
organize components, have a look to the e4 bundles namespace.
Perhaps a useful analogy is with emf and m2m and mdt. Projects
started with a prefix org.eclipse.emf.ocl,
org.eclipse.m2m.qvt.declarative and then lost it; org.eclipse.ocl.
org.eclipse.qvtd... A few well-chosen names keep things
orthogonal. It would be very confusing to have different emf.mtf,
m2m.mtf, mdt.mtf projects so the prefix is not needed. Projects
can be regrouped without renaming. emf and m2m and mdt are
selectively redundant packaging names.
With my suggestion all Impact Analyzer contributions start
org.eclipse.ocl.impact; you're adding a ".tools" and maybe more.
As commented, I want to mainly distinguish between core (non-ui) and
ui (let's no mention "tools" if it's more clear) functionality in
the bundles namespace (as e4 does), as long as we want to package
core (non-ui) in the Core distribution and the UI in the Tools one.
The main reasons:
1. Ease the building and packaging for the releng.
2. Clearly distinct the every Core components (probably run in
standalone) and the Ui one (requires the Eclipse UI platform to
run).
If you prefer the EMF approach rather than the e4 one, it's ok to
me, specially if we can't effectively map the core components at +1
and the ui ones at +3. From the releng point of view, I'm already
managing to package the proper content in the different
distributions, so it's not a problem to me at all.
Finally I would like to better understand
the/understand more inconvenients of my proposal. Your first
argument doesn't suffice to me: we would be suffering of the
same issue if we had had to deal with a separate OCL Tools
project, and the Eclipse "Working Sets" help to organize the
bundles in the workspace (for instance to group bundles
concerning the same feature/tool, regardless its namespace).
If we added a 'proof' component to a separate Tools project we
would not be able to modify the Core project, so if the 'proof'
component has a run-time, we would have to create Tools-runtime
and Tools-ui features to separate them. Users would need to access
one or two of Core/Tool-runtime/Tools-ui. With the merged project,
users just choose one of Core/Tools.
If we had a separate independent OCL Tools project we might be
promoting o.e.o.tools.examples.impact* to o.e.o.tools.impact*.
If we had a separate collaborative OCL Tools project we might be
promoting o.e.o.examples.impact* to o.e.o.impact* which is just
what we have with an integrated Tools project.
A 'proof' component would have its core and ui parts, the design for
that component would imply having bundles under o.e.o.core and
o.e.o.tools/ui
Anyways, to resume... I think that all the alternatives are on the
table. As commented, it's OK to me if your final decision is not
separating the core and ui (what I expect for tools) components via
prefixes.
Regards,
Adolfo.
Regards,
Ed
Adolfo.
El 04/02/2012 11:59, Ed Willink escribió:
Hi Adolfo
Thanks, it's much easier to move forward once you have a stake
in the ground.
Uniform core/tools prefixes seems attractive but, to avoid
confusion from existing facilities, consider what would happen
if we added a new symbolic proof tool.
With uniform core/tools prefixes we might have:
core.proof
core.doc.proof
core.examples.proof
core.tests.proof
tools.proof.compiler
tools.proof.editor
tools.doc.proof.editor
tools.examples.proof.editor
tools.tests.proof.editor
whereas if we group by tool rather than delivery we might have
proof.runtime
proof.compiler
proof.editor
proof.examples
proof.tests
proof.doc
If we need to use examples as an incubation area, during the
examples phase we have no problem with
examples.proof.*
But if we find we need three categories of build rather than
two, using the build category as a package prefix is
unhelpful.
Also, I'm not sure that users want to have core/tools as part
of their imported package path.
For tools at least I think we need a clear tool name as the
first name. Subsequent names we can try to make regular to so
that *.doc.*, *.tests.*, *.examples.*, *.runtime.* makes
releng partitioning fairly easy.
For the current plugins "ecore" and "uml" can be regarded as
tool names, so "ecore.tests" and "uml.tests" are regular
names.
For the Impact Analyzer then "impact" is short and snappy, so
perhaps we want "impact.ecore.runtime",
"impact.ecore.analyzer", "impact.common.analyzer",
"impact.tests", "impact.examples", "impact.doc"
For the Console, I would look towards http://wiki.eclipse.org/MDT/OCL/Debugger
and so we might be looking at "debugger.input",
"debugger.variables", "debugger.breakpoints", ... once we
eventually promote from examples.
For the Xtext editors, "editor..." is probably ok, unless we
want to split as "parser" and "editor". We can decide this
later.
The Pivot model seem the hardest bit, because there is the
neutral domain model. Perhaps "vm.runtime", "vm.model",
"vm.library" so that "pivot" is left with a fairly similar
purpose to "ecore" and "uml".
So we have the following top level 'tools':
common (may not reference any other 'tool')
ecore
uml
pivot
vm
impact
parser
editor
debugger
codegen
and the only plugins that do not comply are org.eclipse.ocl
and org.eclipse.ocl.tests
Tool-specific facilities are in ...tests..., ...examples...,
...doc.... Cross-tool integration is in tests..., examples...,
doc...
The above is for plugin/package names.
I think Core/Tool prefixes would be good for feature names
since if we need to have three deliveries we probably need to
re-arrange features anyway.
Regards
Ed
On 03/02/2012 16:15, Adolfo Sánchez-Barbudo Herrera wrote:
Hi Ed,
Today is quite easy to identify OCL Tools components, that
is:
org.eclipse.ocl.examples.*
The problems come when identifying Core components, unless
we consider Core everything is not o.e.o.examples.*
(However, this doesn't help from the point of view of the
regular expressions which are used, anyway).
From a releng (and any user) perspective it would be quite
easy identify components if we had:
- org.eclipse.ocl.core.* for Core components.
- org.eclipse.ocl.tools.* for Tools components.
This distinction could help, if we desired to have
documentation, examples and tests for both Core and Tools
components:
- org.eclipse.ocl.core.doc for Core API usage
documentation/help.
- org.eclipse.ocl.core.examples.* for Core API usage
examples.
- org.eclipse.ocl.core.tests.* for Core API tests.
- org.eclipse.ocl.tools.doc for Tools documentation/help.
- org.eclipse.ocl.tools.examples.* for Tools examples.
- org.eclipse.ocl.tools.tests.* for (probably UI-based)
tests.
This namespace could also help to identify the future
components of the project. Everything not scoped in
org.eclipse.ocl.core/tools should probably be deprecated and
eventually deleted in future releases. I imagine something
like the following:
org.eclipse.ocl ->
mature, deprecated code
org.eclipse.ocl.uml ->
mature, deprecated code
org.eclipse.ocl.ecore ->
mature, deprecated code
org.eclipse.ocl.ecore.impactanalysis[1] -> impact
analysis for the mature deprecated code
org.eclipse.ocl.core.pivot ->
new pivot implementation
org.eclipse.ocl.core.essentialocl -> new
pivot-based Essential OCL Implementation
org.eclipse.ocl.core.completeocl -> new
pivot-based Complete OCL Implementation
org.eclipse.ocl.core.impactanalysis -> impact
analysis for the new pivot-based implementation
org.eclipse.ocl.core.doc ->
documentation for the Core API usage
org.eclipse.ocl.core.examples.* ->
Examples for Core API usage.
org.eclipse.ocl.core.tests.* ->
tests for Core API.
org.eclipse.ocl.tools.console -> OCL
Console.
org.eclipse.ocl.tools.editor.essentialocl -> Editor for
Essential OCL
org.eclipse.ocl.tools.editor.completeocl -> Editor for
Complete OCL
org.eclipse.ocl.tools.doc ->
documentation/help for Tools.
org.eclipse.ocl.tools.examples.* ->
examples for Tools.
org.eclipse.ocl.tools.tests.* ->
(probably UI-based) tests for Tools.
... and such [2].
Again, this is a suggestion, which could make sense since we
are now distinguising and exposing different Core and Tools
components. It's not a necessity at all.
Adding a new/good features organization, we could finally
distribute Core components in a pure Core Repository, and
the Tools components in a pure Tools Repository. We could
also distribute the current (deprecated and removed in the
future) Core components in a "Deprecated Repository".
Nowadays, because of the current feature organization, the
Core Repository contains only Core Components, however, the
Tools repository contains both Core and Tools components.
Note: [1] Just to name the impact analyzer. This obviously
could be organized as Axel requires/expects.
Note: [2] I know that I'm missing a lot of plugins, but I
hope the idea is caught.
Best Regards,
Adolfo.
El 03/02/2012 14:18, Ed Willink escribió:
Hi Adolfo
Would you like to put together a proposal for revised
feature and plugin names and hierarchy, since your releng
perspective gives you slightly different interests? 4.0.0
is a major version so we can totally reorganize if
absolutely necessary - I hope not.
From a modeling perspective, a Feature contain Features or
Plugins so a simple indented list is sufficient to
identify the intended location of all plugins and features
and Update Site Names.
feature X (Descriptive Name for X)
plugin Y
feature Z (Descriptive Name for Z)
plugin A
For the sake of future proofing, assign names as if all
plugins are promoted from examples now; we'll just defer
renaming examples plugins until they are actually
promoted.
Regards
Ed
On 03/02/2012 13:58, Adolfo Sánchez-Barbudo Herrera wrote:
Finally, I would like to remark (I've not thought about
it) the importance of the new namespace from the point
of view of another stakeholder: The releng :). It could
be interesting to avoid problems like [1] or further
changes in the releng stuff configuration to accomodate
new plugins, having a "coherent" or "uniform" namespace
to distinguish Core components from the Tools one.
[1] https://bugs.eclipse.org/bugs/show_bug.cgi?id=370347#c1
El 03/02/2012 12:03, Ed Willink escribió:
plugin and other global names
These obviously change. The simplest change is just
delete ".examples". Do we want to do something else?
It would be nice if the event plugins went to EMF, but
that doesn't look likely, so they too need review.
It would be nice to have names that can accommodate
the pivot model sometime. I would like to try to
partition the code into the run-time code that
performs (re-)evaluation and the meta-run-time code
that maintains the control objects that make IA so
good. If this is possible, then we want corresponding
names. Perhaps
org.eclipse.ocl.ecore.impact.runtime
org.eclipse.ocl.ecore.impact.analyzer
I hope that migration of the run-time code to align
with the code generated Java can be done quite easily,
since the code generated Java makes no use of any form
of the OCL meta-model; just the polymorphic Values and
polymorphic Domain model for which there is direct and
Reflective Ecore support. Perhaps
org.eclipse.ocl.domain.impact.runtime
Migration of the meta-run-time code will be harder
because that obviously makes use of the OCL
meta-model. Perhaps
org.eclipse.ocl.pivot.impact.analyzer
In order to avoid code duplication, code that is
independent of Ecore/UML/Pivot should be in perhaps
org.eclipse.ocl.common.impact
It may also be appropriate to place some declarations
independent of Ecore/UML/Pivot such as extension
points in
org.eclipse.ocl.common
We cannot easily use org.eclipse.ocl since that is
highly Ecore/UML dependent.
NB being independent of Ecore does not prohibit use of
EObject, EObject.eClass() etc. I hope that the
external API facade can be Ecore/UML/Pivot independent
and so in org.eclipse.ocl.common.impact.
_______________________________________________
mdt-ocl.dev mailing list
mdt-ocl.dev@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/mdt-ocl.dev
_______________________________________________
mdt-ocl.dev mailing list
mdt-ocl.dev@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/mdt-ocl.dev
No
virus found in this message.
Checked by AVG - www.avg.com
Version: 2012.0.1913 / Virus Database: 2112/4781 - Release
Date: 02/02/12
_______________________________________________
mdt-ocl.dev mailing list
mdt-ocl.dev@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/mdt-ocl.dev
_______________________________________________
mdt-ocl.dev mailing list
mdt-ocl.dev@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/mdt-ocl.dev
No virus
found in this message.
Checked by AVG - www.avg.com
Version: 2012.0.1913 / Virus Database: 2112/4792 - Release
Date: 02/06/12
_______________________________________________
mdt-ocl.dev mailing list
mdt-ocl.dev@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/mdt-ocl.dev
|