[squeak-dev] The Trunk: Collections-dtl.560.mcz

David T. Lewis lewis at mail.msen.com
Sun Jan 19 15:55:13 UTC 2014


On Sat, Jan 18, 2014 at 07:29:42PM -0500, David T. Lewis wrote:
> On Fri, Jan 17, 2014 at 11:51:00AM -0500, David T. Lewis wrote:
> > >> How was the source for this version of #workingWithSqueak created? I am
> > >> worried that I might have broken something in the writeScanOn: methods
> > >> for TextAttributes that might have led to generating corrupt source
> > >> code for the workingWithSqueak method ... I hope not :-(
> > >
> > > It was created the normal way -- by simply opening up that workspace,
> > > editing it, and then saving it (Command+s).
> > >
> > > Doing that updates the method source.
> > >
> > > See the acceptBlock in TheWorldMainDockingBar>>#showWelcomeText:label:in:.
> > >
> > 
> > That means that I must have introduced some new bug. I will look for it
> > and provide a fix this weekend.
> 
> Oh my goodness. This appears to be a remarkable confluence of of bugs.
> 
> The welcome text right now just happens to be a Text with a RunArray that
> serializes to a string of exactly 101 characters, including a trailing comma
> that needs to be trimmed. The RunArray>>writeScanOn: method backs the stream
> up one character to trim off that extra comma, leaving the stream at position
> 100. It just so happens that the stream was initialized to size 100, so when
> String>>new:streamContents: checks to see if the stream position matches the
> original size, it sees that the position is 100 so yes they match and it must
> be OK to simply return the #originalContents of the stream instead of going
> to all the extra work of figuring out its current #contents. Unfortunately,
> the stream has actually been advanced to position 101 before being backed
> up to position 100, so its collection has been resized to 200 elements. So
> we grab the 200 elements of the stream collection, assuming that these are
> still the original 100 elements of the original stream collection. That
> gives us a string with an extra comma and 99 null characters at the end
> where we did not expect anything.
> 
> If I load our current #workingWithSqueak into an older Squeak image, the
> above problem is still triggered, and the source code for #workingWithSqueak
> gets corrupted. But it gets corrupted differently than in the latest Squeak,
> presumably due to the changes that I recently made to TextAttributes. And
> in the older Squeak, this mess does not result in an error when opening the
> workingWithSqueak window, probably related to the more permissive parsing
> in the original RunArray class>>scanFrom: parsing.

I put CollectionsTests-dtl.212 in the inbox with a test to document the
issue in OrderedCollectionTest>>testStreamContentsPositioning. I am not
sure how best to handle this, but with respect to the welcome workspace
issue for workingWithSqueak the workaround is to simply add or remove any
text attribute from the workspace text, after which the bug is no longer
triggered.

Dave



More information about the Squeak-dev mailing list