overloaded-box protocol changes

[pnkfelix]

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]

r? @nikomatsakis

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

[pnkfelix]

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

[pnkfelix]

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

[pnkfelix]

(okay I figured out how to accommodate #21988)

[pnkfelix]

cc #22181

[bors]

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

[bors]

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

[bors]

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

[bors]

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

[nikomatsakis]

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

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

[pnkfelix]

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]

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

[bors]

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

[nikomatsakis]

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

[pnkfelix]

@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.