TC39 Editors
2023-12-07
plaintext logs
prev next
source on github
18:13
<bakkot>
writing up the spec for base64 and I really want to use break/continue
18:13
<bakkot>
what if we defined those
18:13
<bakkot>
the alternative is "repeat, while done is false" and then a great deal of nesting
18:18
<Michael Ficarra>
we have early returns, so you could technically use that
18:21
<Michael Ficarra>
also, that reminds me that I wanted to clarify that early returns do, in fact, skip steps that follow
18:51
<ljharb>
a separate AO with early returns seems nice
18:51
<ljharb>
also, that reminds me that I wanted to clarify that early returns do, in fact, skip steps that follow
… how would anyone assume otherwise?
18:55
<Michael Ficarra>
in some (older) programming languages, you could just set a return value at any point in the function (or multiple points) and continue execution
18:56
<Michael Ficarra>
there's no reason to believe that someone reading the spec steps should assume "return" also means "and stop executing further steps"
18:56
<Michael Ficarra>
it could reasonably be understood to mean "set the return value to this"
18:56
<shu>
i think you're confusing setting the return value with a statement saying, control is returned
18:57
<shu>
the older PLs separated the two, but i don't know of any PL that has something that says "return" but doesn't return control
18:58
<bakkot>
restructuring flow to use early returns instead of break/continue can make it worse
18:58
<bakkot>
in this case there is a fair bit of state which would need to be passed into the continuation AO
18:59
<bakkot>
also that doesn't work for continue unless we replace the whole loop with recursion, which would also be worse
19:04
<Michael Ficarra>
the older PLs separated the two, but i don't know of any PL that has something that says "return" but doesn't return control
that's fair, I don't know of any that used the keyword "return" for this
19:05
<Michael Ficarra>
restructuring flow to use early returns instead of break/continue can make it worse
yes I believe this, and was only somewhat serious in my suggestion
19:05
<bakkot>
so, agreed we should add break/continue?
19:05
<Michael Ficarra>
no
19:06
<bakkot>
aww
19:06
<Michael Ficarra>
what do you think about "go to step <emu-xref ..."
19:06
<bakkot>
at least a couple other proposal authors have asked for it
19:06
<bakkot>
uhhhhhhhh
19:06
<Michael Ficarra>
you like it, don't you?
19:06
<bakkot>
not sure goto is better than break/continue but I guess if we carefully don't abuse it it would be ok?
19:07
<Michael Ficarra>
I think I prefer it
19:08
<bakkot>
I think that most proposal authors / readers would be more comfortable with break/continue than goto
19:08
<bakkot>
so can you say more about why you would prefer it?
19:09
<Michael Ficarra>
it's dead simple for a reader
19:10
<bakkot>
so is break/continue, as long as the reader is at all technical
19:10
<Michael Ficarra>
no chance to like incorrectly match it up with the containing repetition or go to the wrong instruction
19:11
<Michael Ficarra>
also, "go to step X" is familiar for any reader
19:12
<bakkot>
yeah but many possible uses are incoherent so when you read it you need to convince yourself it's coherent
19:12
<bakkot>
and the only uses I would want to allow are break/continue
19:12
<bakkot>
so for technical readers, you'd have to manually translate it to break/continue
19:12
<bakkot>
which is dumb
19:12
<bakkot>
we should just write the thing we mean
19:24
<bakkot>
anyway I would be ok with goto instead, it just seems dumb
19:24
<bakkot>
would like to hear from shu I guess
19:30
<ljharb>
with the AO and state passing, would a Record cover that?
19:36
<bakkot>
you still have to construct and pass said record, and record fields are more annoying than local variables, so no, not really
19:54
<shu>
brothers i love goto
19:55
<shu>
jump jump jump jump jump
19:55
<shu>
jump around
19:58
<jmdyck>
@bakkot: would you allow 'break' in For, Repeat, and While ?
20:08
<bakkot>
jmdyck: yes
20:08
<bakkot>
assuming by "while" you mean "Repeat, while"
20:10
<jmdyck>
Actually, there's one step that starts "While", in CleanupFinalizationRegistry. But it doesn't want a break now, and probably won't in future, so probably no loss to leave it out.
20:12
<bakkot>
yeah that's not exactly a loop anyway
20:13
<jmdyck>
With Repeat, there's Repeat-while, Repeat-until, and Repeat-forever
20:16
<jmdyck>
(Repeat-forever currently exits via explicit or implicit 'Return')
20:43
<Richard Gibson>

break/continue, but only labeled

1. [id="step-loop"] Repeat,
  1. Set _i_ to _i_ + 1.
  1. If _i_ ≥ _len_, break out of <emu-xref href="#step-loop"></emu-xref>.
  1. Set _next_ to ? Get(_source_, ! ToString(𝔽(_i_))).
  1. If _next_ is *undefined*, continue to <emu-xref href="#step-loop"></emu-xref>.
  1. …
20:52
<bakkot>
sgtm
20:56
<ljharb>
semantics sound good, but "break out of" would need a pretty clear definition
21:06
<Michael Ficarra>
nah, I still prefer goto
22:00
<jmdyck>
So then it would get rendered as (e.g.) "ii. If i ≥ len, break out of 2.c."
22:02
<jmdyck>
and "iv. If next is undefined, continue to 2.c."
22:07
<jmdyck>
Might want to insert "step".
22:11
<jmdyck>
To simulate a 'break' with 'go to', you'd need to go to the step after the loop, so you'd have to label that step, but then someone might insert a step before that, without realizing that the label needs to be moved. (Similarly if the labelled step needs to move elsewhere.)
22:16
<jmdyck>
To simulate a 'continue' with 'go to', I suppose you'd put a do-nothing step at the end of the body and go to that.
22:34
<Michael Ficarra>
jmdyck: for continue, you can just point to the loop head
22:54
<jmdyck>
That wouldn't be very clear: could suggest restarting the loop.
22:57
<Michael Ficarra>
what does restarting mean?
22:57
<Michael Ficarra>
I don't see how it could possibly be misinterpreted
22:58
<jmdyck>
restarting means execute the step, same as if you go to it from outside.
22:59
<Michael Ficarra>
... that is the correct behaviour?
23:02
<Michael Ficarra>
oh I see, you're thinking of using it with "For ..." loops
23:02
<Michael Ficarra>
I was only considering "repeat"
23:03
<jmdyck>
right.
23:03
<Michael Ficarra>
let's not do that
23:04
<jmdyck>
hah
23:05
<shu>
have i said i love goto
23:11
<jmdyck>
yes, but your intonation is kind of faint