Added a debug flag "-d closure" to explain compilation of
closures (should this be done some other way? Should we
rewrite the "-m" flag to "-d escapes"?)  Used this to
discover that cause was an OXXX node in the captured vars
list, and in turn noticed that OXXX nodes are explicitly
ignored in all other processing of captured variables.

Couldn't figure out a reproducer, did verify that this OXXX
was not caused by an unnamed return value (which is one use
of these).  Verified lack of heap allocation by examining -S
output.

Assembly:
(runtime/mgc.go:1371) PCDATA $0, $2
(runtime/mgc.go:1371) CALL "".notewakeup(SB)
(runtime/mgc.go:1377) LEAQ "".gcBgMarkWorker.func1·f(SB), AX
(runtime/mgc.go:1404) MOVQ AX, (SP)
(runtime/mgc.go:1404) MOVQ "".autotmp_2242+88(SP), CX
(runtime/mgc.go:1404) MOVQ CX, 8(SP)
(runtime/mgc.go:1404) LEAQ go.string."GC worker (idle)"(SB), AX
(runtime/mgc.go:1404) MOVQ AX, 16(SP)
(runtime/mgc.go:1404) MOVQ $16, 24(SP)
(runtime/mgc.go:1404) MOVB $20, 32(SP)
(runtime/mgc.go:1404) MOVQ $0, 40(SP)
(runtime/mgc.go:1404) PCDATA $0, $2
(runtime/mgc.go:1404) CALL "".gopark(SB)

Added a check for compiling_runtime to ensure that this is
caught in the future.  Added a test to test the check.
Verified that 1.5.3 did NOT reject the test case when
compiled with -+ flag, so this is not a recently added bug.

Cause of bug is two-part -- there was no leaking closure
detection ever, and instead it relied on capture-of-variables
to trigger compiling_runtime test, but closures improved in
1.5.3 so that mere capture of a value did not also capture
the variable, which thus allowed closures to escape, as well
as this case where the escape was spurious.  In
fixedbugs/issue14999.go, compare messages for f and g;
1.5.3 would reject g, but not f.  1.4 rejects both because
1.4 heap-allocates parameter x for both.

Fixes #14999.

Change-Id: I40bcdd27056810628e96763a44f2acddd503aee1
Reviewed-on: https://go-review.googlesource.com/21322
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>

Change-Id: Iba82a5bd3846f7ab038cc10ec72ff6bcd2c0b484
Reviewed-on: https://go-review.googlesource.com/21377
Run-TryBot: Dave Cheney <dave@cheney.net>
Reviewed-by: Dave Cheney <dave@cheney.net>

Fix build error on darwin/{arm,arm64} caused by a62ae9f6.

Change-Id: I7da4d147a529b11b3e71352a9276a57595ec4b05
Reviewed-on: https://go-review.googlesource.com/21356
Run-TryBot: Dave Cheney <dave@cheney.net>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

Also cleans up return parameter stutter and missing periods.

Change-Id: I47f5c230227ddfd1b105d5e06842f89ffea50760
Reviewed-on: https://go-review.googlesource.com/21362Reviewed-by: Andrew Gerrand <adg@golang.org>

Issue #8633 (and #9134) noted that we didn't document the rules about
closing the Response.Body when Client.Do returned both a non-nil
*Response and a non-nil error (which can only happen when the user's
CheckRedirect returns an error).

In the process of investigating, I cleaned this code up a bunch, but
no user-visible behavior should have changed, except perhaps some
better error messages in some cases.

It turns out it's always been the case that when a CheckRedirect error
occurs, the Response.Body is already closed. Document that.

And the new code makes that more obvious too.

Fixes #8633

Change-Id: Ibc40cc786ad7fc4e0cf470d66bb559c3b931684d
Reviewed-on: https://go-review.googlesource.com/21364
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Andrew Gerrand <adg@golang.org>

The change in 20907 fixed varexpr but broke aliased.  After that change,
a reference to a field in a struct would not be seen as aliasing itself.
Before that change, it would, but only because all fields in a struct
aliased everything.

This CL changes the compiler to consider all references to a field as
aliasing all other fields in that struct.  This is imperfect--a
reference to one field does not alias another field--but is a simple fix
for the immediate problem.  A better fix would require tracking the
specific fields as well.

Fixes #15042.

Change-Id: I5c95c0dd7b0699e53022fce9bae2e8f50d6d1d04
Reviewed-on: https://go-review.googlesource.com/21390
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>

... when the backend can't be reached.

Fixes #9864

Change-Id: I9e4814824943060e0101b6236d6dfd8d7822b4c7
Reviewed-on: https://go-review.googlesource.com/21363Reviewed-by: Andrew Gerrand <adg@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

Updates x/net/http2 to git rev 31df19d6 for changes since Go 1.6.

The main change was https://go-review.googlesource.com/19726 (move
merging of HEADERS and CONTINUATION into Framer), but there were a few
garbage reduction changes too.

Change-Id: I882443d20749f8638f637a2835efe92538c95d31
Reviewed-on: https://go-review.googlesource.com/21365
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Andrew Gerrand <adg@golang.org>

The default is 10MB, like http2, but can be configured with a new
field http.Transport.MaxResponseHeaderBytes.

Fixes #9115

Change-Id: I01808ac631ce4794ef2b0dfc391ed51cf951ceb1
Reviewed-on: https://go-review.googlesource.com/21329
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Emmanuel Odeke <emm.odeke@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Andrew Gerrand <adg@golang.org>

ANDQConst show up occassionally because of right shifting lowering.
ORs and XORs are already folded properly during generic.

Change-Id: I2f9134679555029c641264ce5333d70e167c65f7
Reviewed-on: https://go-review.googlesource.com/21375Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Alexandru Moșoi <alexandru@mosoi.ro>
TryBot-Result: Gobot Gobot <gobot@golang.org>

Very common, cuts about 70k from pkg/tools/linux_amd64/* binaries.

Change-Id: Ied0c049e56e56a56810c781435d79027fbcaf274
Reviewed-on: https://go-review.googlesource.com/21374Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Alexandru Moșoi <alexandru@mosoi.ro>

In a number of places the code was joining filepaths explicitly with
"/", instead of using filepath.Join. This may cause problems on Windows
(or other) platforms.

This is in support of https://go-review.googlesource.com/#/c/18057

Change-Id: Ieb1334f35ddb2e125be690afcdadff8d7b0ace10
Reviewed-on: https://go-review.googlesource.com/21369Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

Rather than checking the block final bit on the next invocation
of nextBlock, we check it at the termination of the current block.
This ensures that we return (n, io.EOF) instead of (0, io.EOF)
more frequently for most streams.

However, there are certain situations where an eager io.EOF is not done:
1) We previously returned from Read because the write buffer of the internal
dictionary was full, and it just so happens that there is no more data
remaining in the stream.
2) There exists a [non-final, empty, raw block] after all blocks that
actually contain uncompressed data. We cannot return io.EOF eagerly here
since it would break flushing semantics.

Both situations happen infrequently, but it is still important to note that
this change does *not* guarantee that flate will *always* return (n, io.EOF).

Furthermore, this CL makes no changes to the pattern of ReadByte calls
to the underlying io.ByteReader.

Below is the motivation for this change, pulling the text from
@bradfitz's CL/21290:

net/http and other things work better when io.Reader implementations
return (n, io.EOF) at the end, instead of (n, nil) followed by (0,
io.EOF). Both are legal, but the standard library has been moving
towards n+io.EOF.

An investigation of net/http connection re-use in
https://github.com/google/go-github/pull/317 revealed that with gzip
compression + http/1.1 chunking, the net/http package was not
automatically reusing the underlying TCP connections when the final
EOF bytes were already read off the wire. The net/http package only
reuses the connection if the underlying Readers (many of them nested
in this case) all eagerly return io.EOF.

Previous related CLs:
    https://golang.org/cl/76400046 - tls.Reader
    https://golang.org/cl/58240043 - http chunked reader

In addition to net/http, this behavior also helps things like
ioutil.ReadAll (see comments about performance improvements in
https://codereview.appspot.com/49570044)

Updates #14867
Updates google/go-github#317

Change-Id: I637c45552efb561d34b13ed918b73c660f668378
Reviewed-on: https://go-review.googlesource.com/21302
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

Compound AUTO types weren't named previously.  That was because live
variable analysis (plive.go) doesn't handle spilling to compound types.
It can't handle them because there is no valid place to put VARDEFs when
regalloc is spilling compound types.

compound types = multiword builtin types: complex, string, slice, and
interface.

Instead, we split named AUTOs into individual one-word variables.  For
example, a string s gets split into a byte ptr s.ptr and an integer
s.len.  Those two variables can be spilled to / restored from
independently.  As a result, live variable analysis can handle them
because they are one-word objects.

This CL will change how AUTOs are described in DWARF information.
Consider the code:

func f(s string, i int) int {
    x := s[i:i+5]
    g()
    return lookup(x)
}

The old compiler would spill x to two consecutive slots on the stack,
both named x (at offsets 0 and 8).  The new compiler spills the pointer
of x to a slot named x.ptr.  It doesn't spill x.len at all, as it is a
constant (5) and can be rematerialized for the call to lookup.

So compound objects may not be spilled in their entirety, and even if
they are they won't necessarily be contiguous.  Such is the price of
optimization.

Re-enable live variable analysis tests.  One test remains disabled, it
fails because of #14904.

Change-Id: I8ef2b5ab91e43a0d2136bfc231c05d100ec0b801
Reviewed-on: https://go-review.googlesource.com/21233
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>

The conventional name for a sync.Mutex is "mu".

These "lk" names date back to a time before conventions.

Change-Id: Iee57f9f4423d04269e1125b5d82455c453aac26f
Reviewed-on: https://go-review.googlesource.com/21361
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Emmanuel Odeke <emm.odeke@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

For idx1 ops, SP can appear in the index slot.
Swap SP into the base register slot so we can encode
the instruction.

Fixes #15053

Change-Id: I19000cc9d6c86c7611743481e6e2cb78b1ef04eb
Reviewed-on: https://go-review.googlesource.com/21384
Run-TryBot: Keith Randall <khr@golang.org>
Reviewed-by: Alexandru Moșoi <alexandru@mosoi.ro>
TryBot-Result: Gobot Gobot <gobot@golang.org>

Find comparisons to constants and propagate that information
down the dominator tree.  Use it to resolve other constant
comparisons on the same variable.

So if we know x >= 7, then a x > 4 condition must return true.

This change allows us to use "_ = b[7]" hints to eliminate bounds checks.

Fixes #14900

Change-Id: Idbf230bd5b7da43de3ecb48706e21cf01bf812f7
Reviewed-on: https://go-review.googlesource.com/21008Reviewed-by: Alexandru Moșoi <alexandru@mosoi.ro>

Currently we test crc64 only with ISO polynomial.

Change-Id: Ibc5e202db3b960369cbbb18e31eb0fea07b54dba
Reviewed-on: https://go-review.googlesource.com/21309
Run-TryBot: Ilya Tocar <ilya.tocar@intel.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

We already keep the entire pragma bitset in n.Func.Pragma, so there's
no need to track Nointerface separately.

Passes toolstash -cmp.

Change-Id: Ic027ece477fcf63b0c1df128a08b89ef0f34fd58
Reviewed-on: https://go-review.googlesource.com/21381
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

Change-Id: Ib500ee92ae1a3d15f7c9f3f46d238b75184b4304
Reviewed-on: https://go-review.googlesource.com/21382Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

See discussion in #15029.

Change-Id: I5cc8be5737ddb7c1f5e4a6cd92cf557af45e961d
Reviewed-on: https://go-review.googlesource.com/21347Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

Fixes #15026.

Change-Id: I61ed71152b99973270d79264d1e8f466f7343c02
Reviewed-on: https://go-review.googlesource.com/21286
Run-TryBot: David Crawshaw <crawshaw@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>

Add a constant for the magic -1 for slice bounds.
Use it.
Enforce more aggressively that bounds must be
slice, ddd, or non-negative.
Remove ad hoc check in plive.go.
Check bounds before constructing an array type
when typechecking.

All changes are manual.

Passes toolstash -cmp.

Change-Id: I9fd9cc789d7d4b4eea3b30b24037a254d3788add
Reviewed-on: https://go-review.googlesource.com/21348
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>

Pushed from an old client by mistake.  These are the
missing changes.

Change-Id: Ia8d61c5c0bde907369366ea9ea98711823342803
Reviewed-on: https://go-review.googlesource.com/21349Reviewed-by: Todd Neal <todd@tneal.org>

Helpful for indexed loads and stores when the stride is not equal to
the size being loaded/stored.

Update #7927

Change-Id: I8714dd4c7b18a96a611bf5647ee21f753d723945
Reviewed-on: https://go-review.googlesource.com/21346
Run-TryBot: Todd Neal <todd@tneal.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Todd Neal <todd@tneal.org>

We need to make sure all the bounds checks pass before issuing
a load which combines several others.  We do this by issuing the
combined load at the last load's block, where "last" = closest to
the leaf of the dominator tree.

Fixes #15002

Change-Id: I7358116db1e039a072c12c0a73d861f3815d72af
Reviewed-on: https://go-review.googlesource.com/21246Reviewed-by: Todd Neal <todd@tneal.org>

Generated by eg, manually fixed up.

I’m not thrilled about having a setter,
but given the variety of contexts in which this
gets fiddled with, it is the cleanest
available alternative.

Change-Id: Ibdf23e638fe0bdabded014c9e59d557fab8c955f
Reviewed-on: https://go-review.googlesource.com/21341Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>

Finishes cleanup which was too late to do when discovered during the
Go 1.6 cycle.

Fixes #14291

Change-Id: Idc69fadbba10baf246318a22b366709eff088a75
Reviewed-on: https://go-review.googlesource.com/21360
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Andrew Gerrand <adg@golang.org>

Previously, cmd/compile rejected constant int->string conversions if
the integer value did not fit into an "int" value. Also, runtime
incorrectly truncated 64-bit values to 32-bit before checking if
they're a valid Unicode code point. According to the Go spec, both of
these cases should instead yield "\uFFFD".

Fixes #15039.

Change-Id: I3c8a3ad9a0780c0a8dc1911386a523800fec9764
Reviewed-on: https://go-review.googlesource.com/21344Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

Fixes #14610

Change-Id: I3e57dd60b531c1495ea3bc37ef707a1e4e644baa
Reviewed-on: https://go-review.googlesource.com/20180Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

This exports the system cert pool.

The system cert loading was refactored to let it be run multiple times
(so callers get a copy, and can't mutate global state), and also to
not discard errors.

SystemCertPool returns an error on Windows. Maybe it's fixable later,
but so far we haven't used it, since the system verifies TLS.

Fixes #13335

Change-Id: I3dfb4656a373f241bae8529076d24c5f532f113c
Reviewed-on: https://go-review.googlesource.com/21293
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Andrew Gerrand <adg@golang.org>

We create appropriate ELF files automatically based on GOOS. There's
no point in supporting -H elf flag, particularly since we need to emit
different flavors of ELF depending on GOOS anyway.

If that weren't reason enough, -H elf appears to be broken since at
least Go 1.4. At least I wasn't able to find a way to make use of it.

As best I can tell digging through commit history, -H elf is just an
artifact leftover from Plan 9's 6l linker.

Change-Id: I7393caaadbc60107bbd6bc99b976a4f4fe6b5451
Reviewed-on: https://go-review.googlesource.com/21343
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

This only tests amd64 because it's currently broken on non-SSA
backends.

Fixes #8613

Change-Id: I6bc501c81c395e533bb9c7335789750e0c6b7a8f
Reviewed-on: https://go-review.googlesource.com/21325Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

The http2 spec defines a magic string which initates an http2 session:

    "PRI * HTTP/2.0\r\n\r\nSM\r\n\r\n"

It was intentionally chosen to kinda look like an HTTP request, but
just different enough to break things not ready for it. This change
makes Go ready for it.

Notably: Go now accepts the request header (the prefix "PRI *
HTTP/2.0\r\n\r\n") as a valid request, even though it doesn't have a
Host header. But we now mark it as "Connection: close" and teach the
Server to never read a second request from the connection once that's
seen. If the http.Handler wants to deal with the upgrade, it has to
hijack the request, read out the "body", compare it against
"SM\r\n\r\n", and then speak http2. One of the new tests demonstrates
that hijacking.

Fixes #14451
Updates #14141 (h2c)

Change-Id: Ib46142f31c55be7d00c56fa2624ec8a232e00c43
Reviewed-on: https://go-review.googlesource.com/21327Reviewed-by: Andrew Gerrand <adg@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

This makes sure the net/http package never attempts to transmit a
bogus header field key or value and instead fails fast with an error
to the user, rather than relying on the server to maybe return an
error.

It's still possible to use x/net/http2.Transport directly to send
bogus stuff. This change only stops h1 & h2 usage via the net/http
package. A future change will update x/net/http2.

This change also moves some code from request.go to lex.go, which in a
separate future change should be moved so it can be shared with http2
to reduce code bloat.

Updates #14048

Change-Id: I0a44ae1ab357fbfcbe037aa4b5d50669a87f2856
Reviewed-on: https://go-review.googlesource.com/21326Reviewed-by: Andrew Gerrand <adg@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

Test to follow in a separate CL that arranges for the runtime package to
store non-Go addresses in a CPU profile.

Change-Id: I33ce1d66b77340b1e62b54505fc9b1abcec108a9
Reviewed-on: https://go-review.googlesource.com/21055Reviewed-by: Austin Clements <austin@google.com>
Run-TryBot: Ian Lance Taylor <iant@golang.org>

Only use REP;MOVSB if:
 1) The CPUID flag says it is fast, and
 2) The pointers are unaligned
Otherwise, use REP;MOVSQ.

Update #14630

Change-Id: I946b28b87880c08e5eed1ce2945016466c89db66
Reviewed-on: https://go-review.googlesource.com/21300Reviewed-by: Nigel Tao <nigeltao@golang.org>

As a followup to CL 21296, avoid append operations when constructing the
fields of a Type if the length is known beforehand

This also includes some small scoping driveby cleanups, and a change to
tointerface0 to avoid iterating over the field list twice.

compilebench shows a very small reduction in allocations.

 name      old time/op    new time/op    delta
Template     364ms ± 5%     363ms ± 4%    ~     (p=0.945 n=20+19)
Unicode      182ms ±11%     185ms ±12%    ~     (p=0.445 n=20+20)
GoTypes      1.14s ± 2%     1.14s ± 3%    ~     (p=0.221 n=20+20)
Compiler     5.85s ± 2%     5.84s ± 2%    ~     (p=0.369 n=20+20)

name      old alloc/op   new alloc/op   delta
Template    56.7MB ± 0%    56.7MB ± 0%  -0.04%  (p=0.000 n=20+20)
Unicode     38.3MB ± 0%    38.3MB ± 0%    ~     (p=0.728 n=20+19)
GoTypes      180MB ± 0%     180MB ± 0%  -0.02%  (p=0.000 n=20+20)
Compiler     812MB ± 0%     812MB ± 0%  -0.02%  (p=0.000 n=19+20)

name      old allocs/op  new allocs/op  delta
Template      482k ± 0%      480k ± 0%  -0.34%  (p=0.000 n=20+20)
Unicode       377k ± 0%      377k ± 0%  -0.04%  (p=0.010 n=20+20)
GoTypes      1.36M ± 0%     1.35M ± 0%  -0.24%  (p=0.000 n=20+20)
Compiler     5.47M ± 0%     5.46M ± 0%  -0.11%  (p=0.000 n=20+18)

Change-Id: Ibb4c40229fa3816acd8de98ba41d1571a2aabacf
Reviewed-on: https://go-review.googlesource.com/21352Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Dave Cheney <dave@cheney.net>

Apply Robert's optimisation from CL 21241 to Type.Fields. The results
are less impressive, possibly because of the makeup of the test data.

name      old time/op    new time/op    delta
Template     365ms ± 5%     365ms ± 3%    ~     (p=0.888 n=20+16)
Unicode      182ms ±10%     180ms ± 9%    ~     (p=0.883 n=20+20)
GoTypes      1.14s ± 2%     1.13s ± 3%    ~     (p=0.096 n=20+20)
Compiler     5.74s ± 1%     5.76s ± 2%    ~     (p=0.369 n=20+20)

name      old alloc/op   new alloc/op   delta
Template    56.8MB ± 0%    56.7MB ± 0%  -0.15%  (p=0.000 n=19+20)
Unicode     38.3MB ± 0%    38.3MB ± 0%  -0.02%  (p=0.006 n=20+19)
GoTypes      180MB ± 0%     180MB ± 0%  -0.13%  (p=0.000 n=20+20)
Compiler     805MB ± 0%     804MB ± 0%  -0.05%  (p=0.000 n=20+20)

name      old allocs/op  new allocs/op  delta
Template      485k ± 0%      482k ± 0%  -0.54%  (p=0.000 n=19+20)
Unicode       377k ± 0%      377k ± 0%  -0.05%  (p=0.005 n=20+20)
GoTypes      1.37M ± 0%     1.36M ± 0%  -0.53%  (p=0.000 n=20+19)
Compiler     5.42M ± 0%     5.41M ± 0%  -0.21%  (p=0.000 n=20+20)

Change-Id: I6782659fadd605ce9931bf5c737c7058b96a29eb
Reviewed-on: https://go-review.googlesource.com/21296Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

Fixes #14522.

As I said on that issue:

----
This is a progressive JPEG image. There are two dimensions of
progressivity: spectral selection (variables zs and ze in scan.go,
ranging in [0, 63]) and successive approximation (variables ah and al in
scan.go, ranging in [0, 8), from LSB to MSB, although ah=0 implicitly
means ah=8).

For this particular image, there are three components, and the SOS
markers contain this progression:

zs, ze, ah, al:  0  0 0 0	components: 0, 1, 2
zs, ze, ah, al:  1 63 0 0	components: 1
zs, ze, ah, al:  1 63 0 0	components: 2
zs, ze, ah, al:  1 63 0 2	components: 0
zs, ze, ah, al:  1 10 2 1	components: 0
zs, ze, ah, al: 11 63 2 1	components: 0
zs, ze, ah, al:  1 10 1 0	components: 0

The combination of all of these is complete (i.e. spectra 0 to 63 and
bits 8 exclusive to 0) for components 1 and 2, but it is incomplete for
component 0 (the luma component). In particular, there is no data for
component 0, spectra 11 to 63 and bits 1 exclusive to 0.

The image/jpeg code, as of Go 1.6, waits until both dimensions are
complete before performing the de-quantization, IDCT and copy to an
*image.YCbCr. This is the "if zigEnd != blockSize-1 || al != 0 { ...
continue }" code and associated commentary in scan.go.

Almost all progressive JPEG images end up complete in both dimensions
for all components, but this particular image is incomplete for
component 0, so the Go code never writes anything to the Y values of the
resultant *image.YCbCr, which is why the broken output is so dark (but
still looks recognizable in terms of red and blue hues).

My reading of the ITU T.81 JPEG specification (Annex G) doesn't
explicitly say that this is a valid image, but it also doesn't rule it
out.

In any case, the fix is, for progressive JPEG images, to always
reconstruct the decoded blocks (by performing the de-quantization, IDCT
and copy to an *image.YCbCr), regardless of whether or not they end up
complete. Note that, in Go, the jpeg.Decode function does not return
until the entire image is decoded, so we still only want to reconstruct
each block once, not once per SOS (Start Of Scan) marker.
----

A test image was also added, based on video-001.progressive.jpeg. When
decoding that image, inserting a

println("nComp, zs, ze, ah, al:", nComp, zigStart, zigEnd, ah, al)

into decoder.processSOS in scan.go prints:

nComp, zs, ze, ah, al: 3 0 0 0 1
nComp, zs, ze, ah, al: 1 1 5 0 2
nComp, zs, ze, ah, al: 1 1 63 0 1
nComp, zs, ze, ah, al: 1 1 63 0 1
nComp, zs, ze, ah, al: 1 6 63 0 2
nComp, zs, ze, ah, al: 1 1 63 2 1
nComp, zs, ze, ah, al: 3 0 0 1 0
nComp, zs, ze, ah, al: 1 1 63 1 0
nComp, zs, ze, ah, al: 1 1 63 1 0
nComp, zs, ze, ah, al: 1 1 63 1 0

In other words, video-001.progressive.jpeg contains 10 different scans.
This little program below drops half of them (remembering to keep the
"\xff\xd9" End of Image marker):

----
package main

import (
	"bytes"
	"io/ioutil"
	"log"
)

func main() {
	sos := []byte{0xff, 0xda}
	eoi := []byte{0xff, 0xd9}

	src, err := ioutil.ReadFile("video-001.progressive.jpeg")
	if err != nil {
		log.Fatal(err)
	}
	b := bytes.Split(src, sos)
	println(len(b)) // Prints 11.
	dst := bytes.Join(b[:5], sos)
	dst = append(dst, eoi...)
	if err := ioutil.WriteFile("video-001.progressive.truncated.jpeg", dst, 0666); err != nil {
		log.Fatal(err)
	}
}
----

The video-001.progressive.truncated.jpeg was converted to png via
libjpeg and ImageMagick:

djpeg -nosmooth video-001.progressive.truncated.jpeg > tmp.tga
convert tmp.tga video-001.progressive.truncated.png
rm tmp.tga

Change-Id: I72b20cd4fb6746d36d8d4d587f891fb3bc641f84
Reviewed-on: https://go-review.googlesource.com/21062Reviewed-by: Rob Pike <r@golang.org>