Make the bytecode runner return the route that we're going to use. It's
up to the router itself to dispatch to that route.
Besides feeling a teensy bit cleaner, this refactoring is to prepare for
a "Router" middleware, which will allow application developers to
control when in the middleware stack routing occurs.
Previously, a state machine invalidation could have raced against an
in-flight routing attempt: if the invalidation occured after the routing
attempt had already completed its nil-check (choosing not to compile a
new state machine) but before the state machine was atomically loaded to
perform routing, the routing goroutine would begin to panic from
dereferencing nil.
The meat of this change is that we now return the state machine that we
compiled (while still holding the lock), and we only ever interact with
the state machine through atomic pointer loads.
For whatever reason, Go insisted on loading rm.sm[i] in several chunks,
even though it could be loaded in a single 64-bit block. Instead, let's
reorder our loads to minimize the amount of memory we're uselessly
moving around.
This gives us about a 15% perf boost in
github.com/julienschmidt/go-http-routing-benchmark's
BenchmarkGoji_StaticAll, and questionable benefits (i.e., not
distinguishable from noise but certainly no worse) on Goji's own
benchmarks.
The "dryrun" parameter on Pattern.Match was kind of ugly and made for an
exceedingly mediocre public interface. Instead, split its functionality
in two: the previous "dryrun" behavior now lives in the Match method,
and Patterns now actually mutate state when Run is called.
The code on the backend is of course still the same (for now), but at
least the interface is a little nicer.
Previously, the middleware stack passed the router a C, but this was
both odd semantically (a pattern which mutated the environment might see
a *different* environment) and bad for perf: it cost us an allocation.
Now we only pass around *C's internally.
Importantly ("importantly"), this gets us down to 0 allocations for the
static routing case, and one allocation (the URLParams map) for the
normal routing case.
The fast routing diff introduced a regression with how method sets were
calculated for routes that did not match. This fixes that behavior, as
well as making routing considerably more memory-efficient (and therefore
CPU-efficient too) for the case in which many routes share a prefix.
Swap out the naive "try all the routes in order" router with a "compile
a trie down to bytecode" router. It's a ton faster, while providing all
the same semantics.
See the documentation at the top of web/fast_router.go for more.
Partially sort the routes on insertion. We're doing this so we can do
more efficient things to routes later.
The sorting rules are a bit subtle since we aren't allowed to rearrange
routes in a way that would cause the semantics to differ from the dumb
linear scan.
This change replaces a bit of API surface area (the Sub() method on Muxes) with
a slightly more expressive pattern syntax. I'm mostly doing this because it
seems cleaner: the "*" gets to take on a meaning very similar to what it means
in Sinatra (without growing regexp-like middle-of-a-path globbing, which sounds
terrifying and not particularly useful), and we get to nuke a useless function
from the API.
Package web will now add a key to the environment when it fails to find a
valid route for the requested method, but when valid routes exist for other
methods.
This allows either the 404 handler or a sufficiently clever middleware layer to
provide support for OPTIONS automatically.
They say that every programmer builds a web framework at some point. This one is
mine.
The basic idea behind this one is that I wanted a Sinatra for Go, and I couldn't
find one anywhere. Furthermore, net/http is in many ways really close to what I
want out of a Sinatra-in-Go, and many of the frameworks I did find seemed to
reinvent too much, or were incompatible with net/http in weird ways, or used too
much questionable reflection magic. So long story short, I wrote my own.
This implementation is only half-baked, and among other things it's missing a
whole lot of tests.
Carl Jackson
Coda Hale