2009/12/22 Igor Stasenko <siguctua at gmail.com>:
> 2009/12/22 Nicolas Cellier <nicolas.cellier.aka.nice at gmail.com>:
>> 2009/12/22 Igor Stasenko <siguctua at gmail.com>:
>>> 2009/12/22 Nicolas Cellier <nicolas.cellier.aka.nice at gmail.com>:
>>>> Can someone justify why the first sentence would be needed at all:
>>>>    file position: (0 max: position-150).
>>>>    preamble := method getPreambleFrom: file at: (0 max: position - 3).
>>>>
>>>> in ChangeSet>>#scanVersionsOf:class:meta:category:selector:
>>>> and in VersionsBrowser>>#scanVersionsOf:class:meta:category:selector:
>>>>
>>>> My guess is no-one given that the first thing #getPreamble:at: will do
>>>> with the file is
>>>>    file position: (0 max: position - 3).
>>>>
>>>> Just the kind of code I spend too many time on, trying to figure what
>>>> I did not understand when there is nothing to understand but code is
>>>> dead...
>>>> While at it, also explain me why so much code is duplicated in Squeak ;)
>>>>
>>>
>>> Hehe, you forgot
>>> MCMethodDefinition>>scanForPreviousVersion
>>> which is almost identical to the above ones, except some minor changes.
>>> I just stared at same method(s) couple hours ago. :)
>>>
>>> The way how VersionBrowser detects an older versions of method is
>>> total mystery to me.
>>> I doubt that it scans the whole .changes and .sources files, otherwise
>>> it wouldn't be so fast,
>>> or i underestimated the squeak's file handling speed :)
>>>
>>> Oh wait, i think i understood a bit.
>>> it looks for a #methodsFor: prior:  pattern
>>> and by reading a value of prior: argument it knows where to look
>>> further in chain.
>>>
>>> What i would do, if i would be cleaning up this mess, is to change ALL
>>> sourcePointers to point at the beginning
>>> of the method's stamp, instead of pointer to the beginning of method source.
>>> Then, reader could easily read a method stamp/preamble
>>> #nextChunk
>>> and its source
>>> #nextChunk
>>>
>>> and no more black voodoo rewinding the stream back by a 150 ... to
>>> figure out the preamble.
>>>
>>
>> Good idea, but we would have to change not only source position of
>> every method but also prior indication written on each time stamp...
>> I guess a condenseChanges would be the most simple thing to do if
>> you'd go for such a change.
>>
>
> Actually i have better idea in mind, but implementing it could cost a
> lot of blood. :)
> - implement a class SourceCodeManager, which provides interfaces for
> reading/writing sources
> - redirect all sourcePointer-related stuff calls to it.
> - this class should not give away anything other than
>   - source code #source
>   - preamble #preamble
> and most of all, never, never let user code think that it deals with
> file handles.
>

OK, and also deal with opaque prior decoding...

>> Nicolas
>>
>>> And of course, i'm itching to make source pointer to be an arbitrary
>>> integer value - because new method trailers
>>> support that ;)
>>>
>>>
>>>> End of rant :)
>>>>
>>>> Nicolas
>>>>
>>>
>>>
>>> --
>>> Best regards,
>>> Igor Stasenko AKA sig.
>>>
>>>
>>
>>
>
>
>
> --
> Best regards,
> Igor Stasenko AKA sig.
>
>