The FSM used to find matching routes was previously limited to patterns
that contained parameters with the default regexp / no constraints. In
large route sets where many parameters are constrained by custom regexp,
these routes all fall back on a slow linear search over the route list.
These custom regexes were not previously able to be included in the FSM
because it transitioned between nodes using only fragments of the URI,
or path separators [/.?], but a custom regex may cross a path separator
boundary. To work around this, the TransitionTable is improved to
support remembering a point within the matching string that we started,
and continuing to attempt to match from that point up to the end of each
token. Only parameters not on a path separator boundary must still match
with a linear search after this change (e.g. `/foo-:bar/`).
This results in performance for constrainted routes that matches that of
ones using the default regexp.
Benchmark:
https://gist.github.com/theojulienne/e91fc338d180e1710e29c81a5d701fab
Before:
```
Calculating -------------------------------------
without params 6.466k (±12.7%) i/s - 31.648k in 5.009453s
params without constraints
5.867k (±12.9%) i/s - 28.842k in 5.032637s
params with constraints
909.661 (± 7.9%) i/s - 4.536k in 5.023534s
```
After:
```
Calculating -------------------------------------
without params 6.387k (±11.9%) i/s - 31.728k in 5.068939s
params without constraints
5.824k (±13.2%) i/s - 28.650k in 5.043701s
params with constraints
5.406k (±11.7%) i/s - 26.931k in 5.076412s
```
For github.com which has many constrainted parameters, a random sampling
of 10 URL patterns can be matched approximately 2-4x faster than before.
This commit fixes symbols as constrains as tested in
6ab985da28
The FSM used to find matching routes was previously limited to patterns
that contained parameters with the default regexp / no constraints. In
large route sets where many parameters are constrained by custom regexp,
these routes all fall back on a slow linear search over the route list.
These custom regexes were not previously able to be included in the FSM
because it transitioned between nodes using only fragments of the URI,
or path separators [/.?], but a custom regex may cross a path separator
boundary. To work around this, the TransitionTable is improved to
support remembering a point within the matching string that we started,
and continuing to attempt to match from that point up to the end of each
token. Only parameters not on a path separator boundary must still match
with a linear search after this change (e.g. `/foo-:bar/`).
This results in performance for constrainted routes that matches that of
ones using the default regexp.
Benchmark:
https://gist.github.com/theojulienne/e91fc338d180e1710e29c81a5d701fab
Before:
```
Calculating -------------------------------------
without params 6.466k (±12.7%) i/s - 31.648k in 5.009453s
params without constraints
5.867k (±12.9%) i/s - 28.842k in 5.032637s
params with constraints
909.661 (± 7.9%) i/s - 4.536k in 5.023534s
```
After:
```
Calculating -------------------------------------
without params 6.387k (±11.9%) i/s - 31.728k in 5.068939s
params without constraints
5.824k (±13.2%) i/s - 28.650k in 5.043701s
params with constraints
5.406k (±11.7%) i/s - 26.931k in 5.076412s
```
For github.com which has many constrainted parameters, a random sampling
of 10 URL patterns can be matched approximately 2-4x faster than before.
We should build the routes using the user facing API which is `Mapper`.
This frees up the library internals to change as we see fit. IOW we
shouldn't be testing internals.
Partitioning of all the routes is currently being done during the
first request. Since there is no need to clear the cache for
`partitioned_routes` when adding a new route. We can move the
partitioning of the routes during setup time.
Move the Journey code underneath the ActionDispatch namespace so
that we don't pollute the global namespace with names that may
be used for models.
Fixesrails/journey#49.
