Remove for loop in parseSize to enable inlining

[wallyqs]

Using a goto based loop makes it become a leaf function which can be inlined, making us get a slight performance increase as this is in the fast path:

# before
go test ./... -gcflags="-m -m" -run TestParseSize -v 
...
server/util.go:24:6: cannot inline parseSize: unhandled op range

# after
go test ./... -gcflags="-m -m" -run TestParseSize -v
server/util.go:24:6: can inline parseSize as: func([]byte) int {...

Local benchmarks:

benchmark                           old ns/op     new ns/op     delta
BenchmarkParseSize-4                8.81          5.74          -34.85%
Benchmark_____Pub0b_Payload-4       92.9          85.6          -7.86%
Benchmark_____Pub8b_Payload-4       96.1          86.4          -10.09%
Benchmark____Pub32b_Payload-4       106           99.7          -5.94%
Benchmark___Pub128B_Payload-4       131           129           -1.53%
Benchmark___Pub256B_Payload-4       152           143           -5.92%
Benchmark_____Pub1K_Payload-4       348           329           -5.46%
Benchmark_____Pub4K_Payload-4       1474          1440          -2.31%
Benchmark_____Pub8K_Payload-4       3330          3305          -0.75%

benchmark                           old MB/s     new MB/s     speedup
BenchmarkParseSize-4                113.53       174.29       1.54x
Benchmark_____Pub0b_Payload-4       118.45       128.51       1.08x
Benchmark_____Pub8b_Payload-4       197.74       219.83       1.11x
Benchmark____Pub32b_Payload-4       413.47       441.53       1.07x
Benchmark___Pub128B_Payload-4       1072.63      1090.36      1.02x
Benchmark___Pub256B_Payload-4       1765.96      1878.76      1.06x
Benchmark_____Pub1K_Payload-4       2979.37      3150.57      1.06x
Benchmark_____Pub4K_Payload-4       2788.11      2852.61      1.02x
Benchmark_____Pub8K_Payload-4       2464.19      2482.35      1.01x

• Link to issue, e.g. Resolves #NNN
• Documentation added (if applicable)
• Tests added
• Branch rebased on top of current master (git pull --rebase origin master)
• Changes squashed to a single commit (described here)
• Build is green in Travis CI
• You have certified that the contribution is your original work and that you license the work to the project under the MIT license

/cc @nats-io/core

[coveralls]

Coverage decreased (-0.02%) to 91.921% when pulling d19586b on wallyqs:goto-parsesize into b58178d on nats-io:master.

[ghost]

👍

[tylertreat]

lgtm

[kozlovic]

Except for a nitpick, lgtm.

[kozlovic]

nitpick: Could you replace var declarations above with this:

l := len(d)
if l == 0 {
  return -1
}
var (
  i int
  dec byte
)

(assuming that this does not affect performance)

[wallyqs]

Thanks, updated and confirmed we also get a small performance boost:

benchmark                           old ns/op     new ns/op     delta
BenchmarkParseSize-4                8.81          5.73          -34.96%
Benchmark_____Pub0b_Payload-4       92.9          85.0          -8.50%
Benchmark_____Pub8b_Payload-4       96.1          85.4          -11.13%
Benchmark____Pub32b_Payload-4       106           99.3          -6.32%
Benchmark___Pub128B_Payload-4       131           122           -6.87%
Benchmark___Pub256B_Payload-4       152           146           -3.95%
Benchmark_____Pub1K_Payload-4       348           332           -4.60%
Benchmark_____Pub4K_Payload-4       1474          1428          -3.12%
Benchmark_____Pub8K_Payload-4       3330          3257          -2.19%

benchmark                           old MB/s     new MB/s     speedup
BenchmarkParseSize-4                113.53       174.60       1.54x
Benchmark_____Pub0b_Payload-4       118.45       129.37       1.09x
Benchmark_____Pub8b_Payload-4       197.74       222.43       1.12x
Benchmark____Pub32b_Payload-4       413.47       442.96       1.07x
Benchmark___Pub128B_Payload-4       1072.63      1154.81      1.08x
Benchmark___Pub256B_Payload-4       1765.96      1835.92      1.04x
Benchmark_____Pub1K_Payload-4       2979.37      3117.37      1.05x
Benchmark_____Pub4K_Payload-4       2788.11      2877.11      1.03x
Benchmark_____Pub8K_Payload-4       2464.19      2519.17      1.02x

[coveralls]

Coverage decreased (-0.01%) to 91.955% when pulling cd86c99 on wallyqs:goto-parsesize into 0c3d4ce on nats-io:master.

[petemiron]

LGTM.