2019-07-01 [17:45:00.0000] it's fascinating that `CallExpression : CallExpression . IdentifierName` is a separate production from `MemberExpression : MemberExpression . IdentifierName` but evaluated the same (https://tc39.es/ecma262/#sec-property-accessors-runtime-semantics-evaluation) [17:45:35.0000] I would've assumed that `a().b` was a MemberExpression with a CallExpression child [18:06:59.0000] such is the way of mostly-duplicate nonterminal productions [18:07:44.0000] i believe forcing MemberExpression to be its own path is how receivers work [18:10:57.0000] I figure there's a reason but it's quite tricky to identify at a glance [22:19:42.0000] devsnek: i think someone just needs to make a PR/proposal for json modules. [23:03:58.0000] It‘s somewhat involved as a bunch is host-specific still [23:06:06.0000] annevk: how much of that isn't hoistable to 262? [23:06:13.0000] ie what are the tricky parts [23:07:48.0000] MIME type, bytes -> elements, stuff like that [23:08:16.0000] hm, ok [23:08:29.0000] so basically, figuring out where a bunch of host hooks can go? [23:09:14.0000] And keeping it readable [23:09:56.0000] gotcha [23:10:39.0000] Oh, and hosts still need to be able to add yet more types [23:10:53.0000] Seems it’d be rather messy [23:12:02.0000] types like mime types? 2019-07-04 [09:14:06.0000] Oh, https://github.com/tc39/ecma262/pull/1135 got merged! I'd forgotten how nice and precise it makes the spec. [09:14:30.0000] https://tc39.es/ecma262/#sec-mathematical-operations gives me good feelings to read because it's shows such care on the part of the spec to distinguish important, subtle things [09:14:37.0000] Looking forward to adopting this in web specs [09:46:46.0000] Yeah, we blocked formalizing all things numbers on that [12:49:03.0000] this is fancy [14:07:40.0000] caiolima: now that 1135 is merged, it'd be great if you rebased the bigint PR :-) [14:09:35.0000] ljharb: Thx for the heads up! I’ll be able to do this tomorrow or next Monday. [14:11:24.0000] awesome, thanks, lmk 2019-07-05 [14:37:41.0000] ljharb: Maybe remove "good first patch" from https://github.com/tc39/ecma262/issues/1611 (at least for now). It made sense when the change was just removing a clause. Now, it's not so simple. [14:38:47.0000] lol yeah fair 2019-07-06 [17:03:42.0000] leobalter: do you think there's any meaningful overlap between your reverse proposal and my iterator methods proposal? [17:04:07.0000] i think in theory array iterator prototype and map iterator prototype and whatnot could all have a reverse method [22:10:44.0000] devsnek: i think individual iterators would have to opt in to being reversable tho [22:11:08.0000] otherwise it'd be pretty silly to not know if your iterator will be infinite or not (and thus irreversible) [22:11:23.0000] or even just, non deterministically finite [06:53:19.0000] ljharb: I'm more talking about how that integrates with other methods and such [06:58:23.0000] for example in rust you do x.iter().rev(), not x.iter_rev() [07:19:05.0000] devsnek: yeah i think it makes more sense to have a single reverse method than to suffix everything [07:20:18.0000] array.values().reverse() vs array[reverseIterator]() [07:20:53.0000] typically symbol methods are a protocol that nicer-named methods access, no? [07:21:37.0000] perhaps [09:08:59.0000] ljharb, littledan: Hmm, 𝔽 doesn't show up in the default fonts on a Chromebook. :( [09:09:30.0000] (ℝ does, presumably because it's more commonly used as a math symbol) [09:12:10.0000] (I don't have an immediate good suggestion for alternate glyphs, except perhaps ℕ. Misleading in nature, but mnemonic in usage.) [09:13:00.0000] (Or as an amusing alternative, ℚ, because all Numbers are rationals. ^_^) [09:59:49.0000] TabAtkins: isn't 𝔽 usually used for fields anyway? [10:00:24.0000] Yes, so the current usage isn't technically accurate anyway [11:52:17.0000] There was a previous reverse iterator proposal [11:52:29.0000] https://github.com/leebyron/ecmascript-reverse-iterable [11:55:11.0000] oh that does what i said above [11:55:45.0000] People should find the meeting notes where we discussed that [11:56:05.0000] The leebyron proposal looks good to me but maybe there were reasons it didn't advance. (I suspect the biggest reason was lack of champion time.) [11:56:31.0000] oh no https://github.com/tc39/tc39-notes/blob/master/meetings/2015-03/mar-24.md#63-reverseiterable-interface [12:00:08.0000] lollll [12:10:22.0000] TabAtkins: hmm, so it seems like all the double-struck glyphs in Letterlike Symbols block are present in the more "default" fonts like Tinos, Arimo, and Noto Serif, so ℂℍℕℙℚℝℤ work fine. [12:10:40.0000] however, the rest of them are in the newer Mathematical Alphanumeric Symbols block, and are only present in Noto Sans Math, which should also be installed by default on CrOS… [12:12:49.0000] nope, apparently not [14:53:51.0000] rbuckton: what does __promisifiy__ do in ts? [15:06:03.0000] ah nevermind someone else figured out how it's used 2019-07-08 [05:51:28.0000] ljharb: I saw you've commentend on BigInt PR. Thx a lot! I'll be able get back to work on this tomorrow. [08:13:39.0000] devsnek Domenic ljharb I don't have strong preferences how we can add a reverse iterator to the spec, as long as we get something for Array, Map, and Set. Not even worried about String. I'll follow the consent sentiment from the room [08:14:50.0000] the way Lee Byron's proposal describes the iter.reverse() is "safe" but weird, I'm not sure I love it. It should definitely be coordinated with devsnek's proposal. [08:15:37.0000] obviously we need a reverse iteration syntax /s [08:16:32.0000] What I want is something straight to the point where in a list object I can `list.reverse()`. I have a preference (not a blocker) for a string named method, that may mirror a Symbol property. And I'm fully aware Array#reverse is taken, so I'm open to bikeshed names [08:16:57.0000] fwiw i haven't explored optional capabilities at all yet, so iter.reverse might not be the the best way forward, it was just off the top of my head [08:17:37.0000] it's very odd calling `[].values().reverse()` where you get two iterators and rely on both [08:18:41.0000] in rust, when you use rev, what you actually start doing is consuming the same iterator from the back instead of the front [08:18:51.0000] [].valuesRight() is ugly but follow consistency with reduceRight [08:19:01.0000] technically you could consume it from both sides at the same time, and end up in the middle [08:19:52.0000] I'd rather not have a @@reverseIterator depending on a @@iterator [08:20:15.0000] it needn’t have to [08:20:30.0000] it's weird in my mind to be a reverse iterator but not a regular iterator [08:20:31.0000] but that’d be a naive possible implementation [08:20:49.0000] naive possible yes, thats why I'm not discarding [08:20:54.0000] you could also make it required to have an iterator method, but not invoke it [08:20:57.0000] is there an example where you would want to only be iterable with a concept of reverse [08:21:03.0000] I'm playing flexible here [08:21:55.0000] devsnek: you could have an iterable for, say, data points over time, which is infinite - but your reverse iterator could start at a timestamp [08:22:02.0000] devsnek if I have a Map/Set subclass instance that enforces LIFO operations [08:22:22.0000] ok [08:23:35.0000] if we have a concept of double-ended iterators reverse can just be a general wrapper than calls nextBack instead of next [08:23:37.0000] also, the fact these two steps are only applied in specific cases, not applied to generators or any other custom iterator [08:24:24.0000] same for: `[@@asyncIterator]()[@@reverseIterator]()` [08:24:50.0000] or `[@@asyncIterator]().reverse()` where it should not even be available [08:26:16.0000] i'm trying to think of other languages that have this [08:27:01.0000] i think rust and c++ are the only languages that generalize it to a trait/template type thing [08:27:27.0000] and they both use the concept of double-ended [08:45:16.0000] i kinda like this, but perhaps its too complex? https://gc.gy/30305686.png [09:05:04.0000] I think a pro/con list for various possibilities would be a good thing to put in the proposal document [12:12:17.0000] haste [12:12:28.0000] (inb4 jorendorff can say anything posthaste) [12:12:50.0000] > BindingIdentifier: `yield` [12:12:50.0000] > * It is a Syntax Error if this production has a [Yield] parameter. [12:13:06.0000] ^ this appears in the current draft. But how is that different from [12:13:30.0000] > BindingIdentifier: [~Yield] `yield` [12:14:23.0000] That is, seems like an Early Error should be kind of a last resort. We could just enforce this in the grammar. Also, what does "this production" mean? [12:20:31.0000] `yield` in the position of a binding identifier is "this production" [12:27:52.0000] ok, well, that just seems to confirm that the meaning is basically the same as making that production conditional [12:31:14.0000] jorendorff: There is a NOTE above that early error section [12:31:24.0000] "yield and await are permitted as BindingIdentifier in the grammar, and prohibited with static semantics below, to prohibit automatic semicolon insertion in cases such as" [12:32:09.0000] WOW [12:32:11.0000] asi strikes again [12:33:35.0000] :-O [12:34:21.0000] still trying to wrap my head around how that example works though [12:34:37.0000] it's unbelievable, this is the junction of half a dozen JS quirks [12:34:43.0000] `let` is a conditional keyword [12:35:24.0000] `let;` is a legal JS statement in non-strict code; you can do `let = 12; eval("let")` and the answer is 12 [12:37:30.0000] loganfsmyth: So if ASI did happen, you would get `let; await 0;` which would actually parse [12:39:31.0000] ah right, so by making it a valid BindingIdentifier, it's not a syntax error and thus doesn't trigger ASI, ugh [12:39:40.0000] well, not exactly lol [12:39:49.0000] because `let await 0;` *is* still a syntax error [12:39:52.0000] or at least, not a grammar error [12:39:57.0000] yeah, poorly worded [12:40:04.0000] so it *does* trigger ASI [12:40:07.0000] but not until `0` [12:40:30.0000] and ASI looks at the `0`, sees that it's not immediately after a new line, and says no way [12:42:31.0000] so the way this prohibits ASI is by •exploiting details of how ASI case 1 works; and •arranging for a different token to be "the offending token" [12:42:43.0000] yupp [12:42:46.0000] good times :P [12:47:06.0000] loganfsmyth: TIL the Python parser in `python` is generated by this 406-line script https://github.com/python/cpython/blob/master/Parser/pgen/pgen.py [12:47:58.0000] oh cool [12:49:14.0000] apparently python's no-semicolons-required syntax works a *little* differently from JS's [12:51:04.0000] just a tad [12:51:39.0000] ok, i got another one [12:52:13.0000] > TemplateLiteral : NoSubstitutionTemplate [12:52:13.0000] > * It is a Syntax Error if the number of elements in the result of TemplateStrings of TemplateLiteral with argument false is greater than 2^32 - 1. [12:53:04.0000] but TemplateStrings on this is guaranteed to return a List with exactly one element TemplateLiteral:NoSubstitutionTemplate [12:53:31.0000] oops, meant to paste this link https://tc39.es/ecma262/#sec-static-semantics-templatestrings [12:57:40.0000] https://github.com/tc39/ecma262/issues/1588 [12:57:47.0000] jorendorff: ^^ [12:57:52.0000] I think? [12:58:41.0000] yup. thanks [13:49:53.0000] imagine a template literal with 2^32-1 substitutions [13:51:48.0000] devsnek: why just imagine? 😈 2019-07-09 [06:32:11.0000] A syntactic Program breaking that rule would be, I guess, at least 17,179,869,186 characters long. But a test case can be written in <100 characters I'm sure [06:33:02.0000] So ISO dates now support marking some parts of the date with X to indicate unknown things [06:33:35.0000] This is useful for folks that don't have a "complete" date for when they're born or some such [06:33:44.0000] Has this come up for Date? [06:34:05.0000] See https://github.com/whatwg/html/issues/3767 for the HTML side [06:36:43.0000] what is the result of parsing such a date supposed to be [06:38:04.0000] That's a good point, I guess it's somewhat fundamentally incompatible with ES [06:38:27.0000] A number of features in HTML would also do poorly [06:39:32.0000] maybe we can have a function to detect dates with unspecified digits [06:39:55.0000] but trying to create a date object would fail [12:22:18.0000] is it on purpose that GeneratorMethod is missing a production for private fields [12:27:40.0000] oh i guess all the methods are missing [12:27:44.0000] sigh [13:06:15.0000] annevk: it could work with Temporal, i think, for many configurations 2019-07-10 [09:37:37.0000] mathiasbynens: the spec text for Promise.any has `AggregateError` before the introduction, which seems wrong [10:10:40.0000] I thought aggregate error got removed [10:25:30.0000] definitely not 2019-07-11 [08:15:55.0000] Jeez https://bellard.org/quickjs/ is impressive [08:31:53.0000] heh [08:32:04.0000] Bellard's projects are always fun [08:32:34.0000] ..ok, yes, impressive [08:34:33.0000] I was planning to explore BigFloat at some point [08:34:43.0000] cool to see I wasn't the only one [08:43:11.0000] oh this guy made ffmpeg and qemu [09:03:20.0000] yeah, and tcc [09:03:51.0000] and some assorted fun projects, https://bellard.org/jslinux/ e.g. [09:15:10.0000] "this guy made ffmpeg" I NEED TO SHARE WORDS WITH THIS MAN [09:15:35.0000] lol [09:24:30.0000] i've been writing various bash scripts that run ffmpeg for the past several months [09:25:48.0000] and then… https://twitter.com/gesa/status/1148761150529527808 [09:26:23.0000] relatable [09:27:00.0000] a while ago i wrote a dependency resolving build system in js to avoid some mtime bug in make [09:28:16.0000] i *just* wanted a simple postprocessing script for my Plex DVR [09:28:22.0000] and then I met a yak i guess [09:29:09.0000] i feel like all the cool kids these days have plex [09:29:35.0000] "these days" i've had a plex server in my house for a decade [09:30:14.0000] oh wut https://github.com/facebook/hermes [09:30:17.0000] two js engines in one day [10:32:48.0000] wait what: https://github.com/facebook/hermes/blob/master/utils/testsuite/testsuite_blacklist.py#L1462-L1495 [10:33:33.0000] does it expect ES5 as input? [10:34:17.0000] rkirsling: react native is always transpiled [10:34:37.0000] dunno why it would be missing apis though [10:34:58.0000] right I just am surprised it doesn't say so anywhere [10:35:12.0000] actually [10:35:27.0000] right below is "PERMANANT_UNSUPPORTED_FEATURES" [10:35:36.0000] so i guess that stuff will eventually be added [10:36:04.0000] man this thing fails a lot of tests [10:38:03.0000] does facebook participate in ecma at all [10:54:03.0000] devsnek: i haven't seen anyone since 11/2018 [10:54:30.0000] hmm they do appear to be a member [10:55:03.0000] paging efaust 😭 [10:59:15.0000] yeah, wonder where Eric's been lately [12:05:51.0000] he's not on twitter so i can't hassle him 2019-07-12 [07:14:49.0000] mathiasbynens: unsure if bug from jsvu or npm but https://gc.gy/30645865.png [10:34:56.0000] leobalter: are there requirements for test262.report engines or a place to add them? all i can find so far is the issue repo [10:38:12.0000] devsnek we run the results collection privately. In general, we collect the results data using test262-harness, which is powered by eshost [10:38:42.0000] so if something supports eshost [10:38:46.0000] it can be put on test262.report? [10:38:48.0000] yes [10:39:11.0000] that's how we integrated Moddable XS [10:40:15.0000] so how come nashorn and engine262 and such aren't included [10:42:33.0000] think that "progress bar"-esque UI will need to be rethought if we add any more though [10:47:39.0000] probably [10:48:23.0000] i am kinda curious why the report stuff isn't oss [10:49:00.0000] devsnek so far, infrastructure and resources. I'd like to add all of them. I'll forward this internally and see if we can prioritize adding more engines [10:49:56.0000] my answer is wrt adding engines. The matters on being OSS is something I can't answer quickly [10:51:43.0000] thanks! it was more just curiosity than "omg why isn't test262 showing me nashorn's compliance right this minute" [11:00:28.0000] it is probably worth discussing whether there's some sort of "officialness" (probably in the sense of representation in the committee, I guess?) associated with having engines displayed there [11:00:54.0000] maybe so maybe not, just worth having a clear policy on [11:06:11.0000] rkirsling maybe the tests themselves are official, the engines are not specifically "official". We had to pick one engine to start expanding test262.report and used Moddable XS, I'd love to add more engines. [11:17:22.0000] leobalter: yeah I absolutely would like to see more of 'em on there, I just want to make sure we don't lose any ease of comparison :D 2019-07-13 [17:59:40.0000] can anyone make sense of this? https://github.com/Agoric/proposal-infix-bang [18:00:03.0000] deferred operations on objects is what i'm seeing [21:59:17.0000] it looks like a really really large proposal [21:59:34.0000] like, the first one is "promise prototype methods for deferred operations on a fulfillment value" [21:59:40.0000] and the second piece is "syntax for those methods" [22:05:42.0000] ljharb: i think fluent apis can already do the deferred operations tbh [22:05:54.0000] devsnek: what do you mean [22:05:56.0000] like `target!foo` can just be `target.foo` [22:06:03.0000] devsnek: not if `target` is a promise [22:06:07.0000] then it'd be `(await target).foo` [22:06:16.0000] or `target.then(x => x.foo)` really [22:06:26.0000] well your weird proxy thing would be part of a larger api [22:06:36.0000] like in the readme's queue message example [22:06:38.0000] i guess i'm mostly confused why we need syntax for methods nobody's using yet [22:06:45.0000] `messageThing().foo(1, 2, 3)` [22:06:49.0000] can use proxies [22:06:50.0000] it'll be an interesting topic [22:06:52.0000] and thenable [22:07:52.0000] yeah i'm interested to hear the discussion [01:42:24.0000] ljharb: yeah, the methods would need to be added and shown to be widely useful before the syntax could even be considered I’d think [01:48:03.0000] also not sure this needs new syntax at all: I am pretty sure you could just use a proxy which has traps for invocation and property access / assignment, and then use the existing syntax. something like `let remote = makeHandledPromise(); remote.foo(); remote.bar = 42; let value = await settle(remote);` [01:49:46.0000] anyway the point of the proposal, as I understand it, is that you want to have code which represents a sequence of operations on remote data without actually requiring the data to be passed from the remote system back to the system running the code at every step [06:46:09.0000] aside from the new API stuff, it's cool to imagine `fetch()!.json().something` [06:46:36.0000] oops `fetch()!.json()!.something` [06:46:47.0000] perhaps it's a bit easy to forget an exclamation mark [07:33:54.0000] Excited to see the progress on https://github.com/tc39/proposal-explicit-resource-management [07:34:34.0000] I think the infix bang idea is cool but not cool enough to carry its weight in the language. [07:34:50.0000] So many new methods which are just awkward and ugly and nobody would use given that we have await [07:35:06.0000] And the actual bang syntax is pretty error-prone and hard to read [07:35:31.0000] I'd rather explore executing code remotely using something like blöcks [07:50:16.0000] i can't believe we might actually have RAII in the language [07:51:18.0000] i'm kinda sad it's squeezed into the top of a try block [08:01:50.0000] I don't think it counts as RAII if you have to declare a specific using statement [08:01:54.0000] It's more like RAIRA [08:02:22.0000] I.e. the point of RAII is that just intializing is equivalent to resource acquisition [08:02:36.0000] Whereas this requires a specific resource acquisition statement (try, formerly using) [08:05:40.0000] "resource acquisition is resource acquisition" [08:06:16.0000] i just imagine stuff like `try (const _handleScope = v8.handleScope()) {}` [08:06:30.0000] You don't need the extra binding actually [08:06:36.0000] But yes that's the idea [08:06:50.0000] The difference is more in cases like the file handle ones [08:07:08.0000] Whereas in C++ you just initialize the file and poof, resources allocated and deallocated at end of block [08:07:24.0000] In this proposal you need to remember the try() or else you don't get auto-deallocation [08:07:50.0000] (To be clear, this proposal seems to be the best you can do in a GC'ed language.) [08:09:09.0000] could just check every lexical declaration at the end of a block :) [08:10:36.0000] I assume there's a reason that doesn't work but it's too early for me to think hard enough to figure it out [08:10:59.0000] i would assume performance [08:11:14.0000] although i guess engines like v8 already have stuff like "does this object have a toStringTag" [08:11:46.0000] Nah [08:11:50.0000] Google to the rescue https://softwareengineering.stackexchange.com/a/216032/13312 [08:12:05.0000] TLDR it needs a lot more infrastructure like copy constructors, move semantics, etc. [08:13:00.0000] oh yeah [08:13:07.0000] In a language where everything is a reference type, attaching significance to object lifetimes and making things more stack-bound starts crossing into value type territory [08:13:10.0000] hmm [08:13:16.0000] i could always do [08:13:31.0000] `let x; try (const y = whatever) { x = y }` [08:14:00.0000] or i guess just `x = whatever; try (x) {} ` [08:14:09.0000] seems to have the same issue [08:14:16.0000] but i guess its more explicit 2019-07-16 [10:50:24.0000] I randomly just discovered https://github.com/tc39/proposal-optional-chaining exists as a proposal; the overview prose says ?. can't be followed by a decimal digit -- has there been discussion anywhere/yet of this being a breaking change? consider |var increment = twoDecimalDigits?.01:.1;| for example; I couldn't see anything obvious in the repository's issues by their descriptions [10:51:16.0000] jwalden: i think that's why that restriction - because without it, it *would* be a breaking change [10:51:31.0000] ljharb: oh, the restriction means it parses the *other* way [10:51:35.0000] ljharb: okay, fine enough [10:51:48.0000] right [10:51:49.0000] I was assuming the restriction meant that wasn't permitted at all [10:51:59.0000] ah yeah that'd break the web for sure [10:52:11.0000] even if no user has ever typed it that way (which i doubt), minifiers have surely produced it that way [10:52:24.0000] troof [11:09:36.0000] jwalden: just for the record, there's an explanation at the top of the spec draft [11:09:39.0000] > For the syntax, we use the ?. token, with a lookahead at the level of the lexical grammar that allows to discriminate between a?.b (optional chaining) and a?.3:0 (conditional operator, whose meaning cannot be changed due to backward compatibility constraints). [11:09:53.0000] ah [11:12:09.0000] (though of course if the spec text itself implied something different that would be a problem :p) 2019-07-17 [09:03:02.0000] Hrm, why does Math.sqrt(-0) return -0? That's the only argument allowed to be negative, and the only return value that's negative. [09:03:19.0000] Doesn't even make sense, since -0 * -0 = +0 [09:03:39.0000] (Clearly, sqrt(-0) should be 0i.) [09:04:32.0000] (I also wish to again register a complaint with the Math functions test suite, which doesn't have any of these details tested. The spec is clear, but I didn't read into its details when I was copying over the behavior to CSS.) [09:27:08.0000] that is weird [10:50:16.0000] so in the spec, is ℝ the function (or at least applied as function) identical in meaning to ℝ as subscript? I'm looking at https://tc39.es/ecma262/#sec-tostring-applied-to-the-number-type and it talks about ℝ as subscript, but it does not seem to explain ℝ as function at all [10:51:10.0000] is it possible whoever added this was dithering between functional notation and subscript notation, settled on subscript notation ultimately, but missed fixing up a functional use or two, perhaps? [10:53:08.0000] cc ^ caiolima [10:54:31.0000] 5.2.5 (https://tc39.es/ecma262/#sec-mathematical-operations) does have [10:54:31.0000] > A conversion from a Number to a mathematical value is denoted as "the mathematical value of x", or ℝ(x). [10:55:18.0000] seems like it's basically the difference between a literal and a variable then? [10:58:40.0000] (fwiw this was a recent change in https://github.com/tc39/ecma262/pull/1135) [10:59:03.0000] jwalden: I think you mean ℝ(x), the answer is yes. [10:59:19.0000] ah [10:59:51.0000] oh, I missed that in that wall-o-text [11:01:52.0000] ljharb: BTW, not sure if I notified you last week, but I updated https://github.com/tc39/ecma262/pull/1515 since your last review. [11:02:07.0000] thanks! i'll take a look soon 2019-07-18 [09:21:30.0000] In https://tc39.es/ecma262/#sec-createintrinsics, why is thrower created before funcProto [09:21:41.0000] if it was created after, it wouldn't need the SetPrototypeOf call [09:25:40.0000] also test262 expects thrower to have [[Extensible]] = false [09:25:53.0000] but at no point is thrower.[[Extensible]] set to false as far as i can tell [09:27:12.0000] oh the intrinsic says "The value of the [[Extensible]] internal slot of a %ThrowTypeError% function is false." [09:27:23.0000] but the intrinsic is also manually created with a call to CreateBuiltinFunction [09:27:26.0000] and mutated [09:28:54.0000] devsnek: seems like a good editorial improvement :-) [09:29:04.0000] i was just making a commit now [09:34:44.0000] devsnek: I think it *used* to make sense. [09:34:54.0000] I'll see if I can find it. [09:35:28.0000] i can probably remove the entire `AddRestrictedFunctionProperties` thing too [09:37:31.0000] See https://github.com/tc39/ecma262/issues/877 and https://github.com/tc39/ecma262/pull/1148 [09:39:47.0000] hmmmmm [09:39:58.0000] well my thing is purely editorial [09:48:15.0000] ljharb: jmdyck: https://github.com/tc39/ecma262/pull/1635/ [09:48:58.0000] I'll have a look once I'm finished with last night's merges. [11:50:05.0000] `({a, b} = obj);` is a little awkward. I wonder whether do expression could help? Is `do {a, b} = obj;` valid? [11:56:19.0000] Oh, I realize it's a silly question... syntax conflict with current `do {} while` X) [11:57:04.0000] haxjs: https://github.com/tc39/proposal-do-expressions [11:59:27.0000] Thanks, devsnek. I just found my head is not clear in this moment (3AM in my timezone) and decide to go to bed now. '=D [11:59:41.0000] 👍🏻 [12:01:25.0000] Domenic: random question: in the OP of https://github.com/tc39/proposal-weakrefs/issues/17 you assert "To correctly implement this step, you need a reference to every NodeIterator object that's ever been created, so you can update it correctly ". The way I'd expect to implement this is to have the NodeIterator objects, when queried, check the document to see which nodes are still live. is there a reason that implementation doesn't work? [12:02:44.0000] (this doesn't require WeakRefs, and so seems more reasonable) [12:03:15.0000] What do you mean by "when queried"? [12:03:56.0000] oh that reminds me https://gist.github.com/devsnek/e3ee8be1fc235e2bee43b2c1cd262adf [12:04:08.0000] `.nextNode()` etc? [12:04:18.0000] this could be changed to a weakset fairly easily and used in jsdom [12:04:20.0000] or whatever other methods exist for code to look at the contents of a NodeIterator [12:04:36.0000] you have to have a reference to the node iterator first [12:04:52.0000] devsnek: ? [12:05:11.0000] I am proposing that the implementation of the nodeIterator `nextNode` method query the document to see what nodes are still live [12:05:23.0000] if no one has a reference to the nodeIterator, no one is going to call nextNode, and the point is moot [12:06:32.0000] in js though, exists and live are the same thing [12:06:41.0000] until you add weakrefs [12:07:37.0000] devsnek: I have no idea what you are trying to say [12:07:44.0000] maybe i'm misunderstanding [12:07:59.0000] the issue above gives an algorithm for "removing a node" [12:08:08.0000] what I meant by "live" in my first message was "not yet removed" [12:10:03.0000] Bakkot: I am not 100% sure if this was the reason, but I believe you cannot just figure out the next node based on the current state of the tree; you need to see all the intermediate mutations. [12:10:18.0000] So doing it lazily when queried will give wrong results. [12:10:21.0000] ah, fun [13:00:02.0000] Bakkot: Domenic: in particular there’s a reference node affected by mutations [13:01:05.0000] I guess you could use mutation observers [14:10:59.0000] would anyone happen to know of a careful derivation of the length of the longest possible string returned by NumberToString? https://tc39.es/ecma262/#sec-tostring-applied-to-the-number-type the algorithm was not written in a fashion that makes the maximum length clear [15:52:22.0000] jwalden: i don't believe there is a maximum length in the spec; it'd just be memory constraints of the host [15:52:27.0000] jwalden: same as there's no max on string length [15:53:06.0000] tfw `delete a?.b` comes for free in the spec but not the implementation [15:53:07.0000] ljharb: oh there most certainly *is* a maximum length, 64-bit floats mean at most 2**64 possible result strings so they *have* to have an upper limit [15:53:19.0000] jwalden: are bigints 64 bit tho? i thought they were unbounded [15:53:33.0000] jwalden: oh this isn't bigint tho, nvm [15:53:33.0000] ljharb: NumberToString takes in numbers, not bigints [15:59:35.0000] rkirsling: you mean babel? [16:05:51.0000] devsnek: just did a JSC impl :p [16:06:27.0000] rkirsling: why doesn't delete come for free? [16:06:29.0000] we have a hackathon going on so I was like "hmm what do I want to spend a free day on" [16:06:38.0000] doesn't `a?.b` either return undefined or a member reference [16:06:50.0000] dunno, it's possible I maybe a simple mistake, will let you know :D [16:06:55.0000] *made [16:07:09.0000] i'm looking forward to using the impl :) [16:07:41.0000] i added optional chaining to a language i'm working on a few days ago inspired by the js proposal https://github.com/devsnek/slither/commit/09746bc21e2ba0fcb8872af63f4cc62395ba8f03 [16:10:31.0000] awesome [16:11:09.0000] yeah no it looks like (at least for JSC) `delete` makes assumptions about its expression so I have to update them [16:11:41.0000] yeah i would've assumed they have all sorts of shortcuts that make life hard [16:18:18.0000] devsnek: this should give you an idea: https://github.com/WebKit/webkit/blob/master/Source/JavaScriptCore/parser/Nodes.h#L985-L1025 [16:19:16.0000] oh my [16:19:59.0000] that's rough 2019-07-19 [19:36:17.0000] https://tc39.es/ecma262/#table-unicode-script-values seems to look deformed with the `Elym` insertion [19:44:05.0000] mathiasbynens: btw for future updates of Unicode versions, you probably should update the Unicode version in https://tc39.es/ecma262/#sec-normative-references too [22:33:37.0000] TimothyGu: PR? :-p [10:31:29.0000] ljharb: is -0 => 0 in spec prose? [10:31:42.0000] or >=, always forget the order... [10:36:32.0000] I was wondering if that'd matter for https://tc39.es/ecma262/#sec-numbertorawbytes which Atomics.store() ends up invoking, but I'm guessing you get the same either way [11:14:36.0000] annevk: Since the merge of PR #1135, it appears that, in the absence of a subscript, numeric operations in spec prose are interpreted as Number operations. RelationalExpressions use Abstract Relational Comparison, which treats +0 and -0 as the same. [11:21:49.0000] annevk: No, negative zero is treated identically to positive in most contexts. [11:22:12.0000] Whoops skipped over jmdyck's response because it used big words. [11:24:45.0000] (I've been stretching my mouth to let those big words come right out.) [11:43:07.0000] jmdyck: ta 2019-07-20 [17:59:24.0000] there's no example of how to implement evaluating generic module X when you're dynamic importing [17:59:55.0000] shouldn't there be like a `HostEnsureModuleEvaluated(moduleRecord)` that the thing would call or something [18:04:02.0000] and like what do i do if i resolve to a module that is evaluating, the spec says i must return a module that is evaluated, do i wait a tick or something? [18:06:11.0000] if the module isn't cyclic, how would i even know if its been evaluated or not [18:52:29.0000] devsnek: "evaluated, i.e. whose Evaluate concrete method has already been called and returned a normal completion" [18:53:32.0000] Domenic: yes that's the line that is giving me trouble [18:53:55.0000] How your host ensures modules Evaluate() methods have been called is up to your host. [18:54:05.0000] You can see examples of how it's done in browsers by looking at the HTML spec. [18:54:13.0000] i mean for an abstract module record [18:54:17.0000] its safe to just do `Evaluate()` [18:54:24.0000] because its required to return undefined if its already evaluated [18:54:43.0000] Abstract module records Evaluate() doesn't exist [18:54:49.0000] So it is not safe to do it since it doesn't make sense [18:55:08.0000] abstract module records have Link() and Evaluate() [18:55:13.0000] They do not [18:55:17.0000] Concrete subclasses do [18:55:17.0000] test/test262/test/language/expressions/dynamic-import/assignment-expression/unary-expr.js [18:55:23.0000] oops [18:55:26.0000] ok but i mean [18:55:42.0000] the concrete implementations are required to return undefined if its already evaluated [18:56:07.0000] Sure but there will usually be a lot more involved in evaluating a module than just calling Evaluate() [18:56:12.0000] here's the logic i'm using so far https://gc.gy/31292760.png [18:56:24.0000] you'll notice that this can return modules that are `evaluating` or `linking` [18:56:35.0000] i've seen at least one example of it returning a module that is `evaluating` [18:56:38.0000] I suggest studying https://html.spec.whatwg.org/multipage/webappapis.html#hostimportmoduledynamically(referencingscriptormodule,-specifier,-promisecapability) and linked algorithms. [18:56:46.0000] They ensure the result is always evaluated [19:00:37.0000] Domenic: i am not doubting that algorithms to do this can exist [19:00:42.0000] i'm just confused why the spec doesn't provide one [19:01:00.0000] Doing so is intricately entangled in host mechanisms [19:01:55.0000] The spec's job is not to provide a host for finding/linking/evaluating modules. That requires a host. If you want engine262 to be a host, you'll need your own algorithms. The algorithms you use will not generally be useful to other hosts. [19:02:13.0000] its able to run them not-dynamically without any host algorithms [19:02:41.0000]