overloaded-`box` protocol changes #22086

pnkfelix · 2015-02-08T12:32:43Z

This is a bootstrapping illustration of the new box prototype. The main point is to illustrate the fallout that occurs with the compiler type inference as it (nearly) stands today. (Note that much of that fallout has now landed in PR #23002.)

This PR is a revised version of PR #22006, building upon PR #22012 to remove much (but certainly not all) of the fallout observed there.

Now most of the fallout in this PR is from where box <expr> was being used in contexts that are implicitly coerced to Box<Trait>, which is not compatible with the new box protocol (at least, not yet), since the use of coercion there subverts the box-protocol's attempt to infer the appropriate boxed-type to create. (This is noted at this comment on the RFC.)

Here is a self-contained gist, runnable in the playpen, that demonstrates the issue (though note the cfg(works) only actually "works" atop PR Propagate container across object cast #22012), with method names that reflect where I originally found in this in the libsyntax crate.
And here is a second gist that is a bit more narrowly focused on the impl I found of Default for Box<[T]>, which may or may not be an instance of the same problem.

For now, I am usually getting around this by using Box::new(..);

it is worth noting that in HEAP { <expr> } can often fix such cases, but not always;
Box::new(<expr>) always works, though it may introduce an intermediate copy that could have been otherwise avoided.

see also rust-lang/rfcs#809 (which is now merged as of this (updated) writing).

oh, and

[breaking-change]

rust-highfive · 2015-02-08T12:32:54Z

r? @nikomatsakis

(rust_highfive has picked a reviewer for you, use r? to override)

pnkfelix · 2015-02-08T12:55:00Z

(this isn't really up for review, so I removed the assignment to @nikomatsakis , who should nonetheless be cc'ed.)

pnkfelix · 2015-02-08T13:39:11Z

(is not quite right yet, I think I need to accommodate #21988 in some as yet unknown-to-me manner.)

pnkfelix · 2015-02-08T22:10:47Z

(okay I figured out how to accommodate #21988)

pnkfelix · 2015-02-12T19:01:50Z

cc #22181

bors · 2015-03-02T10:08:40Z

☔ The latest upstream changes (presumably #22797) made this pull request unmergeable. Please resolve the merge conflicts.

bors · 2015-03-02T14:31:09Z

☔ The latest upstream changes (presumably #22510) made this pull request unmergeable. Please resolve the merge conflicts.

bors · 2015-03-03T14:18:47Z

☔ The latest upstream changes (presumably #22995) made this pull request unmergeable. Please resolve the merge conflicts.

pnkfelix · 2015-03-06T00:14:59Z

@nikomatsakis (i'm mostly expecting a rubber stamp, though feel free to investigate as much as you like. Oh, and I guess I still owe you some performance and code size comparison data.)

bors · 2015-03-06T08:01:28Z

☔ The latest upstream changes (presumably #22899) made this pull request unmergeable. Please resolve the merge conflicts.

nikomatsakis · 2015-03-06T15:37:50Z

This looks good to me. I'd still like those measurements. ;)

UPDATE: r+, in case it wasn't clear

pnkfelix · 2015-03-11T10:36:55Z

Some rough code size data:

configure --enable-optimize --disable-debug:

Baseline: 6048ba8
Post-runway Pre-protocol: b1ed6333100ac0a1bb46a3ed7aa378b3dfe61d92
Post-protocol: e8f64b6c72442d799df86ff35c8657efba1db44b

stage2 rlib name	Baseline size	Post-runway Pre-protocol size	Post-protocol size	Percent change over Baseline
libcore	16104350	16103898	16114376	(+0.062%)
liblibc	286482	286482	286482	(0%)
librand	830358	830352	830414	(+0.007%)
liballoc	2091636	2091656	`2132392`	(+1.949%)
libunicode	5738940	5738910	5739010	(+0.001%)
libcollections	6116590	6115800	6122776	(+0.101%)
libstd	15271254	15274376	15353768	(+0.540%)
librustc_bitflags	13024	13024	13024	(0%)
libflate	155512	155522	155580	(+0.044%)
libarena	163656	163672	163642	(-0.009%)
liblog	220018	221032	222908	(+1.314%)
libterm	838220	838218	849644	(+1.363%)
libserialize	`3117002`	3117662	3120722	(+0.119%)
libgetopts	658434	657510	659776	(+0.204%)
librbml	654998	655160	655440	(+0.067%)
libtest	1564224	1563860	1579410	(+0.971%)
libgraphviz	155552	155592	155226	(-0.210%)
libfmt_macros	223850	223756	223768	(-0.037%)
libsyntax	31197746	31204328	31303020	(+0.337%)
librustc_llvm	55250354	55250518	55255078	(+0.009%)
librustc_back	1915180	1913664	1920308	(+0.268%)
librustc	40072418	40072128	40018614	(-0.134%)
librustc_typeck	11407138	11409214	11413472	(+0.056%)
librustc_borrowck	2224886	2225584	2232206	(+0.329%)
librustc_resolve	2915826	2915516	2889206	(-0.913%)
librustc_trans	12489192	12492132	12372586	(-0.934%)
librustc_privacy	688020	687596	688842	(+0.119%)
librustc_lint	1398054	1398296	1405288	(+0.517%)
librustc_driver	5246754	5249698	5212736	(-0.648%)
librustdoc	13634096	13640610	13672884	(+0.284%)

total	232643764		232762598	(+0.051%)

Numbers gathered via invocation:

for f in $(ls -tr rust-placer/objdir-opt/x86_64-apple-darwin/stage2/lib/rustlib/x86_64-apple-darwin/lib/stamp.* ) ; do \
   F=$(basename $f | sed -e s/^stamp./lib/ ) ; \
  ( echo $F; \
    ( ls -l rust-plbase/objdir-opt/x86_64-apple-darwin/stage2/lib/rustlib/x86_64-apple-darwin/lib/$F-*.rlib && \
     ls -l rust-placer/objdir-opt/x86_64-apple-darwin/stage2/lib/rustlib/x86_64-apple-darwin/lib/$F-*.rlib )  | \
  cut -f 8 -d ' ' ) | xargs  ; done

(and then manually feeding the data into some scheme code for calculating the percent-deltas)

bors · 2015-03-12T22:46:35Z

☔ The latest upstream changes (presumably #23265) made this pull request unmergeable. Please resolve the merge conflicts.

…ed to Box. Precursor for landing overloaded-`box`, since that will decouple the `box` syntax from the exchange heap (and in fact will eliminate the use of the two aforementioned lang items). Instead, the new demonstration program shows a definition of the `str_eq` lang item. (We do not have that many procedural lang-items to choose from, which is a good sign for our efforts to decouple the compiler from the runtime!) (This previously used a demo of `panic_bounds_check`, but a `str_eq` demonstration is both easier to code and arguably a more interesting aspect of the language to discuss.) Fix unsafe.md example.

See also issue 22405, which tracks going back to `box <expr>` if possible in the future. Precursor for overloaded-`box` and placement-`in`; see Issue 22181.

…ck> }`. Note that `box <expr>` itself remains unchanged.

update test/compile-fail/feature-gate-box-expr.rs to reflect new feature gates. Part of what lands with Issue 22181.

See discussion on Issue 22231.

Namely: * Update run-pass/new-box-syntax * Fix doc-embedded test for `alloc::boxed` to reflect new syntax. * Fix test/debuginfo/box.rs to reflect new syntax. Part of what lands with Issue 22181.

…t suite. Precursor for overloaded-`box` and placement-`in`; see Issue 22181.

Precursor for overloaded-`box` and placement-`in`; see Issue 22181.

bors · 2015-03-13T22:53:06Z

☔ The latest upstream changes (presumably #23292) made this pull request unmergeable. Please resolve the merge conflicts.

nikomatsakis · 2015-04-24T19:45:21Z

@pnkfelix should we close this PR or keep it open?

pnkfelix · 2015-04-25T09:30:18Z

@nikomatsakis i guess i'll close it; I know where to find it when I get a chance to revive it (i.e. after I fix the ptr::write codegen issues.

Hack the move_val_init intrinsic to trans directly into the destination address. This is to remove an intermediate (and unnecessary) alloca on the stack that one otherwise suffers when using this intrinsic. This is part of the `box` protocol work; in particular, this is meant to address the `ptr::write` codegen issues alluded to at this comment: #22086 (comment) cc #22181

rust-highfive assigned nikomatsakis Feb 8, 2015

pnkfelix mentioned this pull request Feb 8, 2015

No checkin: overloaded-box protocol changes #22006

Closed

pnkfelix force-pushed the fsk-box-placer-take-3 branch from 2d0a44d to 0c88f76 Compare February 8, 2015 12:54

pnkfelix unassigned nikomatsakis Feb 8, 2015

pnkfelix force-pushed the fsk-box-placer-take-3 branch 6 times, most recently from 23ea741 to 989515c Compare February 8, 2015 13:12

pnkfelix force-pushed the fsk-box-placer-take-3 branch from b50cab7 to 53062c4 Compare February 8, 2015 22:09

This was referenced Feb 12, 2015

Precedence of box is possibly wrong #21192

Closed

could type inference fill in the _ in expr as Box<_> when expecting Box<Trait> ? #22221

Closed

box/in desugaring leads to inscrutable error messages #22231

Closed

pnkfelix force-pushed the fsk-box-placer-take-3 branch 3 times, most recently from 3430c4d to 2ed35df Compare February 18, 2015 20:54

pnkfelix force-pushed the fsk-box-placer-take-3 branch 3 times, most recently from 9bc1982 to c1d4f12 Compare March 2, 2015 08:01

pnkfelix force-pushed the fsk-box-placer-take-3 branch from c1d4f12 to d0c965d Compare March 2, 2015 14:05

pnkfelix force-pushed the fsk-box-placer-take-3 branch from d0c965d to aaf0563 Compare March 3, 2015 10:14

pnkfelix changed the title ~~No checkin: overloaded-box protocol changes~~ overloaded-box protocol changes Mar 6, 2015

pnkfelix mentioned this pull request Mar 6, 2015

Revise lang_item demo to something unrelated to Box impl #22499

Closed

pnkfelix force-pushed the fsk-box-placer-take-3 branch from 27b06bd to 10f05f9 Compare March 6, 2015 14:32

pnkfelix force-pushed the fsk-box-placer-take-3 branch from 10f05f9 to 52f6957 Compare March 9, 2015 18:56

pnkfelix mentioned this pull request Mar 11, 2015

User-defined placement-box #18233

Closed

pnkfelix force-pushed the fsk-box-placer-take-3 branch from 52f6957 to e8f64b6 Compare March 11, 2015 10:45

pnkfelix added 2 commits March 13, 2015 00:01

Added Box<_> type annotations where necessary for overloaded-box.

f2f09a1

pnkfelix force-pushed the fsk-box-placer-take-3 branch from e8f64b6 to 68b81c1 Compare March 12, 2015 23:02

pnkfelix added 8 commits March 13, 2015 00:03

Switched from box to Box::new where necessary for overloaded-box.

7645a2d

See also issue 22405, which tracks going back to `box <expr>` if possible in the future. Precursor for overloaded-`box` and placement-`in`; see Issue 22181.

parser: switch from box (<place>) <expr> form to `in <place> { <blo…

cbf9303

…ck> }`. Note that `box <expr>` itself remains unchanged.

prototype Placer protocol for unstable overloaded-box and placement-in.

0be597d

Add feature-gates for desugaring-based box and placement-in.

467345b

update test/compile-fail/feature-gate-box-expr.rs to reflect new feature gates. Part of what lands with Issue 22181.

Accommodate error message explosion from box-desugaring in some tests.

4a7df22

See discussion on Issue 22231.

Update tests for desugaring box and placement-in.

056e581

Namely: * Update run-pass/new-box-syntax * Fix doc-embedded test for `alloc::boxed` to reflect new syntax. * Fix test/debuginfo/box.rs to reflect new syntax. Part of what lands with Issue 22181.

Issue 22450: Address desugaring-box problems in [pretty] run-pass tes…

2d44712

…t suite. Precursor for overloaded-`box` and placement-`in`; see Issue 22181.

Workaround issue 22462 by moving static value into its own module.

a0e3c71

Precursor for overloaded-`box` and placement-`in`; see Issue 22181.

pnkfelix force-pushed the fsk-box-placer-take-3 branch from 68b81c1 to a0e3c71 Compare March 12, 2015 23:04

pnkfelix closed this Apr 25, 2015

pnkfelix mentioned this pull request Jun 2, 2015

trans move_val_init expression directly into the destination address. #25959

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

overloaded-`box` protocol changes #22086

overloaded-`box` protocol changes #22086

pnkfelix commented Feb 8, 2015

rust-highfive commented Feb 8, 2015

pnkfelix commented Feb 8, 2015

pnkfelix commented Feb 8, 2015

pnkfelix commented Feb 8, 2015

pnkfelix commented Feb 12, 2015

bors commented Mar 2, 2015

bors commented Mar 2, 2015

bors commented Mar 3, 2015

pnkfelix commented Mar 6, 2015

bors commented Mar 6, 2015

nikomatsakis commented Mar 6, 2015

pnkfelix commented Mar 11, 2015

bors commented Mar 12, 2015

bors commented Mar 13, 2015

nikomatsakis commented Apr 24, 2015

pnkfelix commented Apr 25, 2015

overloaded-box protocol changes #22086

overloaded-box protocol changes #22086

Conversation

pnkfelix commented Feb 8, 2015

rust-highfive commented Feb 8, 2015

pnkfelix commented Feb 8, 2015

pnkfelix commented Feb 8, 2015

pnkfelix commented Feb 8, 2015

pnkfelix commented Feb 12, 2015

bors commented Mar 2, 2015

bors commented Mar 2, 2015

bors commented Mar 3, 2015

pnkfelix commented Mar 6, 2015

bors commented Mar 6, 2015

nikomatsakis commented Mar 6, 2015

pnkfelix commented Mar 11, 2015

bors commented Mar 12, 2015

bors commented Mar 13, 2015

nikomatsakis commented Apr 24, 2015

pnkfelix commented Apr 25, 2015

overloaded-`box` protocol changes #22086

overloaded-`box` protocol changes #22086