Add commentary to explain better what's going on, but the
code change is a simple one-line reversal to the previous
form.

R=rsc
CC=golang-dev
https://golang.org/cl/6428072

R=nigeltao
CC=golang-dev
https://golang.org/cl/6431055

Don't unescape entities in attributes when they don't end with
a semicolon and they are followed by '=', a letter, or a digit.

Pass 6 more tests from the WebKit test suite, plus one that was
commented out in token_test.go.

R=nigeltao
CC=golang-dev
https://golang.org/cl/6405073

R=golang-dev, bradfitz
CC=golang-dev
https://golang.org/cl/6422056

R=golang-dev, dsymonds
CC=golang-dev
https://golang.org/cl/6432062

R=golang-dev, r
CC=golang-dev
https://golang.org/cl/6434043

R=golang-dev, dave, r
CC=golang-dev
https://golang.org/cl/6430062

R=golang-dev, r, iant
CC=golang-dev
https://golang.org/cl/6416056

Fixes #3844.

R=golang-dev, r
CC=golang-dev
https://golang.org/cl/6432054

R=golang-dev, bsiegert, r
CC=golang-dev
https://golang.org/cl/6428050

Surrogates are still admitted, but I have sent mail to golang-dev on that topic.

Fixes #3785.

R=golang-dev, rogpeppe, iant
CC=golang-dev
https://golang.org/cl/6398049

Fixes #3839.

R=golang-dev, bradfitz
CC=golang-dev
https://golang.org/cl/6428053

If a Client's CheckRedirect function returns an error, we
again return both a non-nil *Response and a non-nil error.

Fixes #3795

R=golang-dev, n13m3y3r
CC=golang-dev
https://golang.org/cl/6429044

Fixes #3830.

R=golang-dev, r
CC=golang-dev, remy
https://golang.org/cl/6373047

By default, all are still run, but a particular test can be
selected with the new flags.

R=golang-dev, bradfitz
CC=golang-dev
https://golang.org/cl/6395053

Fixes #3822.

R=golang-dev, r, dsymonds, minux.ma, adg
CC=golang-dev
https://golang.org/cl/6395043

R=golang-dev, bradfitz
CC=golang-dev
https://golang.org/cl/6392065

Fixes #3815.

R=r
CC=golang-dev
https://golang.org/cl/6352109

R=golang-dev, bradfitz
CC=golang-dev
https://golang.org/cl/6399047

Some tests currently fail with GOMAXPROCS>1

R=golang-dev
CC=golang-dev
https://golang.org/cl/6398044

Some class of bugs (data races, runtime bugs) can be found
only with real parallelism.
Note that GOMAXPROCS=32 is somewhat different from go test -cpu=32,
this intentionally uses GOMAXPROCS to stress program bootstrap,
testing code, garbage collections, etc.
Package selection is mostly random.

R=golang-dev, dave, r
CC=golang-dev
https://golang.org/cl/6346070

They can generate huge amounts of memory, causing failure on
small machines. Also they can be very slow. So slow that one test
was commented out! We uncomment it and use a flag.

Fixes #3742.

R=golang-dev, bradfitz
CC=golang-dev
https://golang.org/cl/6373044

- Changed the representation of colElem to support a few cases
  for some languages not supported by the current format.
- Changed offsets for implicit primary values. This makes the
  values both easier to read and debug (last 4 nibbles are identical to
  implicit primary value) and also results in better packing.
- Fixed bug in weight conversion code that did not pop up yet by
  sheer luck.
Note that tables.go also includes changes to the contraction trie
from CL 6346092.

R=r, mpvl
CC=golang-dev
https://golang.org/cl/6392060

which has a rather large contraction table. The value of the next state
offset now starts after the current block, instead of before.  This is
slightly less efficient (on extra addition per state change), but gives
some extra range for the offsets.
Also introduced constants for final (0) and noIndex (0xFF).
tables.go is updated in a separate CL.

R=r
CC=golang-dev
https://golang.org/cl/6346092

Fixes #3334.

R=golang-dev, r
CC=golang-dev, remy
https://golang.org/cl/6350103

Gob decoding reads a whole message into memory and then
copies it into a bytes.Buffer. For large messages this wastes
an entire copy of the message. In this CL, we use a staging
buffer to avoid the large temporary.

Update #2539
RSS drops to 775MB from 1GB.
Active memory drops to 858317048 from 1027878136,
essentially the size of one copy of the input file.

R=dsymonds, nigeltao
CC=golang-dev
https://golang.org/cl/6392057

Allows a client to pre-allocate buffer space that is known to be necessary,
avoiding expensive reallocations.

R=gri, gri, adg
CC=golang-dev
https://golang.org/cl/6392061

The error was caused by a call to implements() even when
the type switch variable was not an interface.

Fixes #3786.

R=golang-dev, r
CC=golang-dev, remy
https://golang.org/cl/6354102

Fixes #3818.

R=golang-dev, rsc, r
CC=golang-dev, remy
https://golang.org/cl/6352106

benchmark           old ns/op    new ns/op    delta
BenchmarkHilbert      6253043      6267289   +0.23%
BenchmarkMul         45355940     39490633  -12.93%

R=r
CC=golang-dev
https://golang.org/cl/6355104

Reuse temporary slice to avoid extra allocations
(originally done correctly by remyoudompheng@gmail.com
in https://golang.org/cl/6345075/).

benchmark           old ns/op    new ns/op    delta
BenchmarkHilbert      6252790      6262304   +0.15%
BenchmarkMul         45827438     45301002   -1.15%

R=r
CC=golang-dev
https://golang.org/cl/6346097

R=golang-dev, dsymonds, r, nigeltao
CC=golang-dev
https://golang.org/cl/6374059

The previous implementation used to have a O(n) recursion
depth for unbalanced inputs. A test is added to check that a
reasonable amount of bytes is allocated in this case.

Fixes #3807.

R=golang-dev, dsymonds, gri
CC=golang-dev, remy
https://golang.org/cl/6345075

This is the remainder of https://golang.org/cl/4601051.

Partially addresses issue 2705.

R=golang-dev, r, bradfitz, minux.ma
CC=golang-dev
https://golang.org/cl/6354066

Also adds a new test for GOMAXPROCS=16 explicitly, which now passes
reliably in a stress loop like:

$ go test -c
$ (while ./http.test -test.v -test.run=Concurrency; do echo pass; done ) 2>&1 | tee foo; less foo

(It used to fail very quickly and reliably on at least Linux/amd64)

Fixes #3793

R=golang-dev, adg, r
CC=golang-dev
https://golang.org/cl/6347061

R=golang-dev, dave, r
CC=golang-dev, remy
https://golang.org/cl/6378046

OpenSSL requires that RSA signatures be exactly the same byte-length
as the modulus. Currently it'll reject ~1/256 of our signatures: those
that end up a byte shorter.

Fixes #3796.

R=golang-dev, edsrzf, r
CC=golang-dev
https://golang.org/cl/6352093

Fixes #3424.

R=fullung, adg
CC=golang-dev
https://golang.org/cl/6343088

Previously, strings that didn't have an explicit ASN.1 string type
were taken to be ASN.1 PrintableStrings. This resulted in an error if
a unrepresentable charactor was included.

For compatibility reasons, I'm too afraid to switch the default string
type to UTF8String, but this patch causes untyped strings to become
UTF8Strings if they contain a charactor that's not valid in a
PrintableString.

Fixes #3791.

R=golang-dev, bradfitz, r, r
CC=golang-dev
https://golang.org/cl/6348074

Entrust have issued a root certificate that's not marked as valid for
signing certificates.

This results in Go programs failing to validate certificates that
chain up to this root (i.e. gateway.push.apple.com:2195).

Although this is clearly a mistake on Entrust's part, it seems that we
will have to bodge around it.

R=golang-dev, dave, r
CC=golang-dev
https://golang.org/cl/6346064