On Mon, Apr 30, 2018 at 02:13:15PM -0700, Frank Shearar wrote:
On 30 April 2018 at 13:23, David T. Lewis lewis@mail.msen.com wrote:
On Fri, Apr 27, 2018 at 10:36:17PM -0500, Tim Johnson wrote:
Hi,
I ran some tests and found a couple that are failing. I also checked and found these tests don't fail in my 5.1 image.
BrowserTest>>#testBuildMessageCategoryBrowserEditString DebuggerExtensionsTest>>#testCollectionsGeneralise
I think the first one might actually be a case where a bug was fixed. The test fails because of a timeout, because there is a dialog on-screen which is not returning. Not sure, but the test may be intended to declare that the dialog should appear, and now it is (?).
The second one, I don't understand. There is no comment.
I don't understand it either, and strangely it has no method timestamp. But the test was was introduced in April 2013 in this update:
Name: ToolsTests-fbs.62 Author: fbs Time: 19 April 2013, 8:43:40.116 am UUID: 926d563e-d57b-44ec-b4e7-672010293c2b Ancestors: ToolsTests-nice.61
Tests for the new #canonicalArgumentName Debugger methods.
This is part of test coverage for #canonicalArgumentName, which is implemented in Object and also has no method timestamp or comment (!!!).
The canonicalArgumentName method is sent by many unit tests, so it has very good coverage (even though I don't know the significance of this failing test).
Aside from unit tests, it is sent by Message>>createStubMethod and appears to be used when the debugger provides a template for implementing a missing method, or for implementing a method override.
The test does this:
testCollectionsGeneralise self assert: Collection name equals: Array new canonicalArgumentName. self assert: Collection name equals: OrderedCollection new canonicalArgumentName. self assert: Collection name equals: LinkedList new canonicalArgumentName
This looks like a regression that should be fixed such that the tests pass again, and that also deserves a good method comment in Object>>cononicalArgumentName.
I think that some more background and explanation can probably be found on squeak-dev circa April 2013, notably this reference:
http://lists.squeakfoundation.org/pipermail/squeak-dev/2013- April/170506.html
Which points to this:
Name: Tools-fbs.460 Author: fbs Time: 19 April 2013, 8:40:24.143 am UUID: d5cf82c4-bda7-48ff-bfbd-e8b27d0a07d7 Ancestors: Tools-fbs.459
When creating a stub method, give the argument names that represent the (usual) desired name more accurately. For instance, Arrays, OrderedCollections and Sets all result in 'aCollection', ByteStrings and WideStrings in 'aString', and so on.
So perhaps that last paragraph about "when creating a stub method..." might serve as a method comment for Object>>cononicalArgumentName?
Dave
It was part of my work to improve the "JIT development" workflow, aka "debugger driven development". See http://lists.squeakfoundation.org/pipermail/squeak-dev/2013-April/170693.htm... and http://bugs.squeak.org/view.php?id=7761
Thanks Frank,
And kudos for the test coverage, it is the sort of thing that almost certainly would have gone unnoticed otherwise.
Checking what has changed, the following two additions to the image account for the test failure:
ArrayedCollection>>canonicalArgumentName ==> 'Array' SequenceableCollection>>canonicalArgumentName ==> 'Sequence'
These entered the image here:
Name: Tools-eem.788 Author: eem Time: 6 January 2018, 3:37:50.088654 pm UUID: bb90e476-4cf4-47bd-a8be-bc2785cc8504 Ancestors: Tools-eem.787
Add some more canonicalArgumentName implementations for well-known Collection subclasses.
So apparently the right thing to do is to update the unit tests to reflect these additions.
Dave