Re: [Pharo-project] we need help for rome

Previous Topic Next Topic
 
classic Classic list List threaded Threaded
8 messages Options
Reply | Threaded
Open this post in threaded view
|

Re: [Pharo-project] we need help for rome

Bert Freudenberg

I think only the first 4 inst vars are actually used by the plugin:

http://squeakvm.org/svn/squeak/trunk/platforms/unix/src/plugins/RomePlugin/RomePlugin.c
#define CanvasHandleIndex 0
#define CanvasTargetIndex 1
#define CanvasFlagsIndex 2
#define CanvasStrokeColorIndex 3
#define CanvasInstSize 8

So on Linux, Rome does use the "right" size. John appears to have used a Sophie image to generate his Plugin, which makes this fail (though I could not find his Rome plugin in the Mac OS platform code so can't verify that theory).

The proper fix would be to modify RomePlugin class>>initializeInstVarIndices to not generate the indices, but simply hard-code only the used ones (setting CanvasInstSize to 4), and change the comparison John mentioned below to <=. Also remove the unused *Index class vars. Now that there are plugins in the wild, hard-coding the indices is a good idea anyway. During development, the flexible generator version was useful, but not anymore.

- Bert -

On 18.04.2010, at 18:45, Stéphane Ducasse wrote:

>
> John
>
> how can I know the order and the iv that are used by the plugin?
>
> Stef
>
> On Apr 17, 2010, at 4:09 AM, John M McIntosh wrote:
>
>> (a) In Sophie  
>> RomeCanvas subclass: #RomePluginCanvas
>> instanceVariableNames: 'handle target flags strokeColor stack font fontSize fontMatrix currentFillColor currentBackColor glyphAccuracy cachedGlyphBlt cachedClearBlt drawRunPositions drawRunGlyph drawRunPointY'
>> classVariableNames: 'CachedTarget FlagFill FlagStro
>>
>> (b) In Pharo
>>
>> RomeCanvas subclass: #RomePluginCanvas
>> instanceVariableNames: 'handle target flags strokeColor stack font fontSize fontMatrix'
>> classVariableNames: 'FlagFill FlagStroke Registry'
>> poolDictionaries: ''
>> category: 'Rome-PluginCanvas'
>>
>> Comment warning:
>>
>> INST VAR ORDER IS KNOWN TO PLUGIN! DO NOT REARRANGE!
>>
>>
>> So in the plugin we have...
>>
>> if (interpreterProxy->slotSizeOf(canvasOop)) < CanvasInstSize)
>> fail().
>>
>>
>> where CanvasInstSize is 13
>>
>> but as you see in (b) the number of instance slots for the canvas Oops is  8.
>>
>> 8 < 13, oops you fail.
>>
>> So where did the instance vars   currentFillColor currentBackColor glyphAccuracy cachedGlyphBlt cachedClearBlt
>> go? On purpose gone, refactored, old code. etc....
>>
>>
>> On 2010-04-16, at 1:37 PM, Stéphane Ducasse wrote:
>>
>>>
>>> On Apr 16, 2010, at 10:30 PM, John McIntosh wrote:
>>>
>>>> Well I'm just as puzzled as you since the primitive gets executed and you are using the same binary that was shipped with Sophie years back.
>>>
>>> I like to hear that because I feel less idiot. :)

Reply | Threaded
Open this post in threaded view
|

Re: [Pharo-project] we need help for rome

stephane ducasse-2

Thanks Bert

>
> I think only the first 4 inst vars are actually used by the plugin:
>
> http://squeakvm.org/svn/squeak/trunk/platforms/unix/src/plugins/RomePlugin/RomePlugin.c
> #define CanvasHandleIndex 0
> #define CanvasTargetIndex 1
> #define CanvasFlagsIndex 2
> #define CanvasStrokeColorIndex 3
> #define CanvasInstSize 8

how did you get that information? from the image side or C code?
Because I tried to see but unsure.

>
> So on Linux, Rome does use the "right" size. John appears to have used a Sophie image to generate his Plugin, which makes this fail (though I could not find his Rome plugin in the Mac OS platform code so can't verify that theory).

Now I understand why alain could remove the instance variable when cleaning the image side.


> The proper fix would be to modify RomePlugin class>>initializeInstVarIndices to not generate the indices, but simply hard-code only the used ones (setting CanvasInstSize to 4), and change the comparison John mentioned below to <=. Also remove the unused *Index class vars. Now that there are plugins in the wild, hard-coding the indices is a good idea anyway. During development, the flexible generator version was useful, but not anymore.


I browsed a bit the C code and I'm learning so may be my questions are not clever, but

        - do we need the pango primitives for ROME?
        Apparently Pango may use Cairo for fonts rendering but do Rome needs that?
        So is it simpler to use fonts via pango than cairo.

        - then what about FormInstSize?

        (interpreterProxy->slotSizeOf(formOop)) < FormInstSize
        does it mean
       
        - the primitives we have access to are listed by
                void* RomePlugin_exports[][3] = {
       
        - how do we know that the plugin is somehow in sync with Cairo "latest" version?

Stef

>
> - Bert -
>
> On 18.04.2010, at 18:45, Stéphane Ducasse wrote:
>>
>> John
>>
>> how can I know the order and the iv that are used by the plugin?
>>
>> Stef
>>
>> On Apr 17, 2010, at 4:09 AM, John M McIntosh wrote:
>>
>>> (a) In Sophie  
>>> RomeCanvas subclass: #RomePluginCanvas
>>> instanceVariableNames: 'handle target flags strokeColor stack font fontSize fontMatrix currentFillColor currentBackColor glyphAccuracy cachedGlyphBlt cachedClearBlt drawRunPositions drawRunGlyph drawRunPointY'
>>> classVariableNames: 'CachedTarget FlagFill FlagStro
>>>
>>> (b) In Pharo
>>>
>>> RomeCanvas subclass: #RomePluginCanvas
>>> instanceVariableNames: 'handle target flags strokeColor stack font fontSize fontMatrix'
>>> classVariableNames: 'FlagFill FlagStroke Registry'
>>> poolDictionaries: ''
>>> category: 'Rome-PluginCanvas'
>>>
>>> Comment warning:
>>>
>>> INST VAR ORDER IS KNOWN TO PLUGIN! DO NOT REARRANGE!
>>>
>>>
>>> So in the plugin we have...
>>>
>>> if (interpreterProxy->slotSizeOf(canvasOop)) < CanvasInstSize)
>>> fail().
>>>
>>>
>>> where CanvasInstSize is 13
>>>
>>> but as you see in (b) the number of instance slots for the canvas Oops is  8.
>>>
>>> 8 < 13, oops you fail.
>>>
>>> So where did the instance vars   currentFillColor currentBackColor glyphAccuracy cachedGlyphBlt cachedClearBlt
>>> go? On purpose gone, refactored, old code. etc....
>>>
>>>
>>> On 2010-04-16, at 1:37 PM, Stéphane Ducasse wrote:
>>>
>>>>
>>>> On Apr 16, 2010, at 10:30 PM, John McIntosh wrote:
>>>>
>>>>> Well I'm just as puzzled as you since the primitive gets executed and you are using the same binary that was shipped with Sophie years back.
>>>>
>>>> I like to hear that because I feel less idiot. :)
>

Reply | Threaded
Open this post in threaded view
|

Re: [Pharo-project] we need help for rome

johnmci
In reply to this post by Bert Freudenberg
 
Well hey I'm just the guy tapping the compile button, er no support agreement here...

Likely the original plugin code came via

MCHttpRepository
        location: 'http://source.impara.de/Rome'
        user: ''
        password: ''

If someone wants to make the required fixes to the plugin code, and the smalltalk code I *think* I can build a new plugin..  Likely this won't affect Sophie users (if any) because that's a one click app. However this means an update to the mac vm's and other platforms in order to ensure the smalltalk class in a Pharo image matches expectations.

On 2010-04-18, at 10:45 AM, Bert Freudenberg wrote:

>
> I think only the first 4 inst vars are actually used by the plugin:
>
> http://squeakvm.org/svn/squeak/trunk/platforms/unix/src/plugins/RomePlugin/RomePlugin.c
> #define CanvasHandleIndex 0
> #define CanvasTargetIndex 1
> #define CanvasFlagsIndex 2
> #define CanvasStrokeColorIndex 3
> #define CanvasInstSize 8
>
> So on Linux, Rome does use the "right" size. John appears to have used a Sophie image to generate his Plugin, which makes this fail (though I could not find his Rome plugin in the Mac OS platform code so can't verify that theory).
>
> The proper fix would be to modify RomePlugin class>>initializeInstVarIndices to not generate the indices, but simply hard-code only the used ones (setting CanvasInstSize to 4), and change the comparison John mentioned below to <=. Also remove the unused *Index class vars. Now that there are plugins in the wild, hard-coding the indices is a good idea anyway. During development, the flexible generator version was useful, but not anymore.
>
> - Bert -
>
--
===========================================================================
John M. McIntosh <[hidden email]>   Twitter:  squeaker68882
Corporate Smalltalk Consulting Ltd.  http://www.smalltalkconsulting.com
===========================================================================





smime.p7s (3K) Download Attachment
Reply | Threaded
Open this post in threaded view
|

Re: [Pharo-project] we need help for rome

Bert Freudenberg

Well, the Linux one is fine, as I wrote. It is built from

        http://www.squeaksource.com/Rome.html

You could simply use that to build the plugin, which should work fine.

Or someone could merge the Sophie stuff from the impara repo, adding the fix I suggested.

The impara repos are read-only and not used anymore AFAIK, so I'd suggest committing to the squeaksource repo.

- Bert -

On 18.04.2010, at 23:23, John M McIntosh wrote:

>
> Well hey I'm just the guy tapping the compile button, er no support agreement here...
>
> Likely the original plugin code came via
>
> MCHttpRepository
> location: 'http://source.impara.de/Rome'
> user: ''
> password: ''
>
> If someone wants to make the required fixes to the plugin code, and the smalltalk code I *think* I can build a new plugin..  Likely this won't affect Sophie users (if any) because that's a one click app. However this means an update to the mac vm's and other platforms in order to ensure the smalltalk class in a Pharo image matches expectations.
>
> On 2010-04-18, at 10:45 AM, Bert Freudenberg wrote:
>
>>
>> I think only the first 4 inst vars are actually used by the plugin:
>>
>> http://squeakvm.org/svn/squeak/trunk/platforms/unix/src/plugins/RomePlugin/RomePlugin.c
>> #define CanvasHandleIndex 0
>> #define CanvasTargetIndex 1
>> #define CanvasFlagsIndex 2
>> #define CanvasStrokeColorIndex 3
>> #define CanvasInstSize 8
>>
>> So on Linux, Rome does use the "right" size. John appears to have used a Sophie image to generate his Plugin, which makes this fail (though I could not find his Rome plugin in the Mac OS platform code so can't verify that theory).
>>
>> The proper fix would be to modify RomePlugin class>>initializeInstVarIndices to not generate the indices, but simply hard-code only the used ones (setting CanvasInstSize to 4), and change the comparison John mentioned below to <=. Also remove the unused *Index class vars. Now that there are plugins in the wild, hard-coding the indices is a good idea anyway. During development, the flexible generator version was useful, but not anymore.
>>
>> - Bert -
>>
>
> --
> ===========================================================================
> John M. McIntosh <[hidden email]>   Twitter:  squeaker68882
> Corporate Smalltalk Consulting Ltd.  http://www.smalltalkconsulting.com
> ===========================================================================
>
>
>
>


Reply | Threaded
Open this post in threaded view
|

Re: [Pharo-project] we need help for rome

Bert Freudenberg
In reply to this post by stephane ducasse-2

On 18.04.2010, at 20:14, stephane ducasse wrote:

>
>
> Thanks Bert
>
>>
>> I think only the first 4 inst vars are actually used by the plugin:
>>
>> http://squeakvm.org/svn/squeak/trunk/platforms/unix/src/plugins/RomePlugin/RomePlugin.c
>> #define CanvasHandleIndex 0
>> #define CanvasTargetIndex 1
>> #define CanvasFlagsIndex 2
>> #define CanvasStrokeColorIndex 3
>> #define CanvasInstSize 8
>
> how did you get that information? from the image side or C code?
> Because I tried to see but unsure.

It's in the C code I linked above. I just reordered it for better readability.

>> So on Linux, Rome does use the "right" size. John appears to have used a Sophie image to generate his Plugin, which makes this fail (though I could not find his Rome plugin in the Mac OS platform code so can't verify that theory).
>
> Now I understand why alain could remove the instance variable when cleaning the image side.
>
>> The proper fix would be to modify RomePlugin class>>initializeInstVarIndices to not generate the indices, but simply hard-code only the used ones (setting CanvasInstSize to 4), and change the comparison John mentioned below to <=. Also remove the unused *Index class vars. Now that there are plugins in the wild, hard-coding the indices is a good idea anyway. During development, the flexible generator version was useful, but not anymore.
>
>
> I browsed a bit the C code and I'm learning so may be my questions are not clever, but
>
> - do we need the pango primitives for ROME?
> Apparently Pango may use Cairo for fonts rendering but do Rome needs that?
> So is it simpler to use fonts via pango than cairo.

Pango provides word/paragraph layout. Many non-latin scripts (e.g. Devanagari) require "glyph shaping", there is no one-to-one mapping between Unicode characters and glyphs rendered in context. Pango can do that. I'm not aware of any other renderer in Squeak that would support this (Yoshiki added it for OLPC). Well, maybe the Unicode plugin from Scratch could do it too, not sure.

> - then what about FormInstSize?
>
> (interpreterProxy->slotSizeOf(formOop)) < FormInstSize
> does it mean

Same thing. The instance variables actually used by the plugin should be hard coded. Otherwise the same would happen as in Sophie - some inst vars were added to the Canvas, so the inst size compiled into the plugin became larger than necessary.

Btw, by now you probably have guessed that a workaround for the Mac issue would be to add 5 dummy inst vars to RomePluginCanvas just to make the plugin happy.

> - the primitives we have access to are listed by
> void* RomePlugin_exports[][3] = {

Right.

> - how do we know that the plugin is somehow in sync with Cairo "latest" version?

It doesn't need to be "in sync". But maybe I do not understand the question.

- Bert -

> Stef
>
>>
>> - Bert -
>>
>> On 18.04.2010, at 18:45, Stéphane Ducasse wrote:
>>>
>>> John
>>>
>>> how can I know the order and the iv that are used by the plugin?
>>>
>>> Stef
>>>
>>> On Apr 17, 2010, at 4:09 AM, John M McIntosh wrote:
>>>
>>>> (a) In Sophie  
>>>> RomeCanvas subclass: #RomePluginCanvas
>>>> instanceVariableNames: 'handle target flags strokeColor stack font fontSize fontMatrix currentFillColor currentBackColor glyphAccuracy cachedGlyphBlt cachedClearBlt drawRunPositions drawRunGlyph drawRunPointY'
>>>> classVariableNames: 'CachedTarget FlagFill FlagStro
>>>>
>>>> (b) In Pharo
>>>>
>>>> RomeCanvas subclass: #RomePluginCanvas
>>>> instanceVariableNames: 'handle target flags strokeColor stack font fontSize fontMatrix'
>>>> classVariableNames: 'FlagFill FlagStroke Registry'
>>>> poolDictionaries: ''
>>>> category: 'Rome-PluginCanvas'
>>>>
>>>> Comment warning:
>>>>
>>>> INST VAR ORDER IS KNOWN TO PLUGIN! DO NOT REARRANGE!
>>>>
>>>>
>>>> So in the plugin we have...
>>>>
>>>> if (interpreterProxy->slotSizeOf(canvasOop)) < CanvasInstSize)
>>>> fail().
>>>>
>>>>
>>>> where CanvasInstSize is 13
>>>>
>>>> but as you see in (b) the number of instance slots for the canvas Oops is  8.
>>>>
>>>> 8 < 13, oops you fail.
>>>>
>>>> So where did the instance vars   currentFillColor currentBackColor glyphAccuracy cachedGlyphBlt cachedClearBlt
>>>> go? On purpose gone, refactored, old code. etc....
>>>>
>>>>
>>>> On 2010-04-16, at 1:37 PM, Stéphane Ducasse wrote:
>>>>
>>>>>
>>>>> On Apr 16, 2010, at 10:30 PM, John McIntosh wrote:
>>>>>
>>>>>> Well I'm just as puzzled as you since the primitive gets executed and you are using the same binary that was shipped with Sophie years back.
>>>>>
>>>>> I like to hear that because I feel less idiot. :)
>>
>


Reply | Threaded
Open this post in threaded view
|

Re: [Pharo-project] we need help for rome

stephane ducasse-2
In reply to this post by johnmci

Ok I will ask cyrille if he can have a look at the c :)

Stef

On Apr 18, 2010, at 11:23 PM, John M McIntosh wrote:

> Well hey I'm just the guy tapping the compile button, er no support agreement here...
>
> Likely the original plugin code came via
>
> MCHttpRepository
> location: 'http://source.impara.de/Rome'
> user: ''
> password: ''

This is strange because I think that alain took it from there too.
May be not.

Stef


>
> If someone wants to make the required fixes to the plugin code, and the smalltalk code I *think* I can build a new plugin..  Likely this won't affect Sophie users (if any) because that's a one click app. However this means an update to the mac vm's and other platforms in order to ensure the smalltalk class in a Pharo image matches expectations.
>
> On 2010-04-18, at 10:45 AM, Bert Freudenberg wrote:
>
>>
>> I think only the first 4 inst vars are actually used by the plugin:
>>
>> http://squeakvm.org/svn/squeak/trunk/platforms/unix/src/plugins/RomePlugin/RomePlugin.c
>> #define CanvasHandleIndex 0
>> #define CanvasTargetIndex 1
>> #define CanvasFlagsIndex 2
>> #define CanvasStrokeColorIndex 3
>> #define CanvasInstSize 8
>>
>> So on Linux, Rome does use the "right" size. John appears to have used a Sophie image to generate his Plugin, which makes this fail (though I could not find his Rome plugin in the Mac OS platform code so can't verify that theory).
>>
>> The proper fix would be to modify RomePlugin class>>initializeInstVarIndices to not generate the indices, but simply hard-code only the used ones (setting CanvasInstSize to 4), and change the comparison John mentioned below to <=. Also remove the unused *Index class vars. Now that there are plugins in the wild, hard-coding the indices is a good idea anyway. During development, the flexible generator version was useful, but not anymore.
>>
>> - Bert -
>>
>
> --
> ===========================================================================
> John M. McIntosh <[hidden email]>   Twitter:  squeaker68882
> Corporate Smalltalk Consulting Ltd.  http://www.smalltalkconsulting.com
> ===========================================================================
>
>
>
>

Reply | Threaded
Open this post in threaded view
|

Re: [Pharo-project] [Vm-dev] Re: we need help for rome

Bert Freudenberg
In reply to this post by Bert Freudenberg

On 19.04.2010, at 10:42, Stéphane Ducasse wrote:

>
>
> On Apr 18, 2010, at 11:57 PM, Bert Freudenberg wrote:
>
>> On 18.04.2010, at 20:14, stephane ducasse wrote:
>>>
>>>
>>> Thanks Bert
>>>
>>>>
>>>> I think only the first 4 inst vars are actually used by the plugin:
>>>>
>>>> http://squeakvm.org/svn/squeak/trunk/platforms/unix/src/plugins/RomePlugin/RomePlugin.c
>>>> #define CanvasHandleIndex 0
>>>> #define CanvasTargetIndex 1
>>>> #define CanvasFlagsIndex 2
>>>> #define CanvasStrokeColorIndex 3
>>>> #define CanvasInstSize 8
>>>
>>> how did you get that information? from the image side or C code?
>>> Because I tried to see but unsure.
>>
>> It's in the C code I linked above. I just reordered it for better readability.
>
> Yes I saw my question was more how do we identify code that is not used anymore?
> We just do the closure over the published primitives and the rest means that it is useful.
> My question was how did you identify that only 4 inst were used and not more.

If you look at the RomePlugin class you see that 8 class vars are defined. But in the generated C code, only the first four are declared. That is because the Slang translator only exports the used ones. That's how I know the others are unused - they were not exported :)

>>>> The proper fix would be to modify RomePlugin class>>initializeInstVarIndices to not generate the indices, but simply hard-code only the used ones (setting CanvasInstSize to 4), and change the comparison John mentioned below to <=. Also remove the unused *Index class vars. Now that there are plugins in the wild, hard-coding the indices is a good idea anyway. During development, the flexible generator version was useful, but not anymore.
>
>
> Ok I will browse the RomePlugin class with Cyrille and Jean-Baptiste
>
>
>>> I browsed a bit the C code and I'm learning so may be my questions are not clever, but
>>>
>>> - do we need the pango primitives for ROME?
>>> Apparently Pango may use Cairo for fonts rendering but do Rome needs that?
>>> So is it simpler to use fonts via pango than cairo.
>>
>> Pango provides word/paragraph layout. Many non-latin scripts (e.g. Devanagari) require "glyph shaping", there is no one-to-one mapping between Unicode characters and glyphs rendered in context. Pango can do that. I'm not aware of any other renderer in Squeak that would support this (Yoshiki added it for OLPC). Well, maybe the Unicode plugin from Scratch could do it too, not sure.
>>
>>> - then what about FormInstSize?
>>>
>>> (interpreterProxy->slotSizeOf(formOop)) < FormInstSize
>>> does it mean
>>
>> Same thing. The instance variables actually used by the plugin should be hard coded. Otherwise the same would happen as in Sophie - some inst vars were added to the Canvas, so the inst size compiled into the plugin became larger than necessary.
>>
>> Btw, by now you probably have guessed that a workaround for the Mac issue would be to add 5 dummy inst vars to RomePluginCanvas just to make the plugin happy.
>
> Yes. In fact my question was more is FormInstSize used (or are the functionality using FormInstSize used)?

If you see it in the C code then yes, it is used.

> I have problems to see how we can clean the C code.
> All my questions should be interpreted with this slant.

Well, no need to "clean the C code" because it is just generated. Clean the RomePlugin class instead :^)

>>> It doesn't need to be "in sync". But maybe I do not understand the question.
>
> Sorry was not clear.
> My question is how Cairo supports backwards compatibility?
> What happens if the plugin expect a different version of Cairo on the machine?
> May be they do not change their interface?

Many projects are better about backwards compatibility than we are ;)

Generally, once an API was added, it is *never* removed or changed in the same major version of a library. Only additions are allowed in point releases. This ensures code can dynamically link to older binaries and not break with newer releases.

Also, the major version is part of the library name (executables actually link against bla.so.1 not bla.so) so multiple major versions of the same library can co-exist on a system. So major releases can remove or change existing API.

In short, I wouldn't worry about it - I'm not sure in cairo's particular case but in general that's how it works.

There is something to be said about published vs. public interfaces, and in Smalltalk we usually don't respect either ...

- Bert -

Reply | Threaded
Open this post in threaded view
|

Re: [Pharo-project] [Vm-dev] Re: we need help for rome

stephane ducasse-2

>>>
>>
>> Yes I saw my question was more how do we identify code that is not used anymore?
>> We just do the closure over the published primitives and the rest means that it is useful.
>> My question was how did you identify that only 4 inst were used and not more.
>
> If you look at the RomePlugin class you see that 8 class vars are defined. But in the generated C code, only the first four are declared. That is because the Slang translator only exports the used ones. That's how I know the others are unused - they were not exported :)


Ok this is what I wanted to know :)

>>>>>
> Well, no need to "clean the C code" because it is just generated. Clean the RomePlugin class instead :^)

Excellent.

>>>>
> Many projects are better about backwards compatibility than we are ;)
>
> Generally, once an API was added, it is *never* removed or changed in the same major version of a library. Only additions are allowed in point releases. This ensures code can dynamically link to older binaries and not break with newer releases.
>
> Also, the major version is part of the library name (executables actually link against bla.so.1 not bla.so) so multiple major versions of the same library can co-exist on a system. So major releases can remove or change existing API.
>
> In short, I wouldn't worry about it - I'm not sure in cairo's particular case but in general that's how it works.

Excellent!

> There is something to be said about published vs. public interfaces, and in Smalltalk we usually don't respect either ...

:)

Stef

>
> - Bert -
>