Hi Christoph --
Do I understand correctly that you do not yet consider this good enough? :)
That depends on your original motivation. :-) If you want to closure a binding from an outer context, there is no need for this inline optimization in the first place.
Having a one-arg block in "otherwise:" only makes sense for dynamic block re-use:
explorer := [:obj | obj explore]. ... a caseOf: cases otherwise: explorer. ...
Why would I want to rename an existing binding in the static case?
a := 42. a caseOf: cases otherwise: [a explore]. "Why [:x | x explore]?" ...
Best, Marcel Am 22.11.2021 17:18:57 schrieb christoph.thiede@student.hpi.uni-potsdam.de christoph.thiede@student.hpi.uni-potsdam.de: Hi Levente, Hi Marcel,
currently, this expression:
self assert: 42 equals: (6 caseOf: { [a] -> [b]. [b] -> [c]. [c] -> [a] } otherwise: [:x | x * 7]).
is decompiled like that:
self assert: 42 equals: ((x := 6) caseOf: { [a] -> [b]. [b] -> [c]. [c] -> [a]} otherwise: [x * 7])
x is also declared correctly in the test method. Do I understand correctly that you do not yet consider this good enough? :)
Best, Christoph
--- Sent from Squeak Inbox Talk [https://github.com/hpi-swa-lab/squeak-inbox-talk]
On 2021-11-18T14:04:16+01:00, marcel.taeumel@hpi.de wrote:
Hi Levente --
Thanks. Decompiler only adds the block-arg for not-inlined caseOfs. See ObjectTest >> #testCaseOfOtherwise.
Hmm... Christoph? =) Otherwise (ha!) I will take a look at it soon-ish.
Best, Marcel Am 18.11.2021 13:51:02 schrieb Levente Uzonyi <leves at caesar.elte.hu>: Hi Marcel,
I just checked how this works, and noticed that the decompiler cannot recreate the original code if the otherwise: block has an argument.
Levente
On Thu, 18 Nov 2021, Marcel Taeumel wrote:
Merged.
Am 21.02.2020 17:33:03 schrieb Thiede, Christoph :
This changeset makes it possible to accept an argument in the otherwise block of a #caseOf:otherwise: call.
Diff of the #caseOf:otherwise: implementation:
caseOf: aBlockAssociationCollection otherwise: aBlock
"The elements of aBlockAssociationCollection are associations between blocks. Answer the evaluated value of the first association in aBlockAssociationCollection whose evaluated key equals the receiver. If no match is found, answer the result of evaluating aBlock."
aBlockAssociationCollection associationsDo:
[:assoc | (assoc key value = self) ifTrue: [^assoc value value]].
- ^ aBlock value
- ^ aBlock cull: self
"| z | z := {[#a]->[1+1]. ['b' asSymbol]->[2+2]. [#c]->[3+3]}. #b caseOf: z otherwise: [0]"
"| z | z := {[#a]->[1+1]. ['d' asSymbol]->[2+2]. [#c]->[3+3]}. #b caseOf: z otherwise: [0]"
"The following are compiled in-line:"
"#b caseOf: {[#a]->[1+1]. ['b' asSymbol]->[2+2]. [#c]->[3+3]} otherwise: [0]"
"#b caseOf: {[#a]->[1+1]. ['d' asSymbol]->[2+2]. [#c]->[3+3]} otherwise: [0]"
+ "#b caseOf: {[#a]->[1+1]. ['d' asSymbol]->[2+2]. [#c]->[3+3]} otherwise: [:x | x halt]"
Furthermore, the changeset includes a necessary modification of MessageNode >> #transformCase: so that the otherwise argument can be compiled in-line.
Last but not least, I wrote some tests for #caseOf:[otherwise:].
Please review!
(In a later change, it would be possible to allow arguments for the association key blocks as well. But I love short feedback loops, so let's assess this one first :-))
Best,
Christoph