Clojurians Log v2

See the release notes for 0.7.0, this one macro was moved to its own library.

For me the biggest weakness with pedestal routing is how it handles path parameters. Both in terms of expressiveness (1), but also complexity (2). 1. This is discussed here: https://tonsky.me/blog/simple-router/ For 2) the guide says this: > The map tree router is by far the fastest router in Pedestal. Part of how it gets that speed, though, is by forbidding the use of dynamic path segments like wildcards and path parameters. > If your routes include any of them, then even if you request the map tree router, Pedestal will fall back to using the (still pretty fast) prefix tree router. That seems odd to me, because it sounds like an unnecessary performance cliff. If none of my first path segments are dynamic, shouldn't I benefit from the map speed for dealing with that part of the routing? I would think only matching the dynamic parts of the routing tree would need to use a slower alternative.

That's a good point, and that would be a terrific optimization for the new router (which I call sawtooth, because that's the name that popped into my head).

Sawtooth works by subdividing the routes by each route's port, host, scheme, and method; after that it is routing on the routes' :path. It's essentially breaking the routing table into many little shards, and some of those shards (or subselections of those shards) may be free of wildcards, and could instead be based on a map lookup.

Like a lot of my efforts, the point of sawtooth is to move as much of the logic into a one-time compile phase, to minimize work done at runtime.

That sounds awesome

prefix-tree does some of that, but at the end, it has a bucket of possible routes and has to do some ugly work to find the "best match".

The code is in the hls/842-conflicting-routes branch.

Reading Tonsky's notes ... once I have Sawtooth working and it's basic value proven, then we can figure out how to, for example, allow "/user/create" to match in preference to "/user/:id" if desired.

Enthusiastic about composable routing tables. But not about /user/create shadowing /user/*. Of course I need not design routes that overlap... But in my generosity and selflessness, I think only of the convenience of others others when I gripe and fulminate that this is the sort of thing that people should joyfully program in their own apps and then not blame the framework when Charlie Reate asks why his user account doesn't work.

Yay, found the bug I introduced into prefix-tree router, now I can start comparing perfix-tree and sawtooth head-to-head.

Initial results are ok, but not great:

┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━━┓
┃           Expression           ┃    Mean   ┃     Var    ┃    Ratio    ┃
┣━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╋━━━━━━━━━━━╋━━━━━━━━━━━━╋━━━━━━━━━━━━━┫
┃  (execute :small :prefix-tree) ┃ 102.94 µs ┃  ± 9.90 µs ┃     100.0 % ┃ (fastest)
┃     (execute :small :sawtooth) ┃ 109.25 µs ┃  ± 4.89 µs ┃     106.1 % ┃
┃ (execute :medium :prefix-tree) ┃ 943.94 µs ┃ ± 23.98 µs ┃     917.0 % ┃
┃    (execute :medium :sawtooth) ┃   1.13 ms ┃ ± 33.57 µs ┃   1,101.4 % ┃
┃  (execute :large :prefix-tree) ┃  94.68 ms ┃  ± 1.20 ms ┃  91,975.2 % ┃
┃     (execute :large :sawtooth) ┃ 113.50 ms ┃  ± 2.55 ms ┃ 110,254.1 % ┃ (slowest)
┗━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┻━━━━━━━━━━━┻━━━━━━━━━━━━┻━━━━━━━━━━━━━┛

I thought the up-front costs of sawtooth would be offset by the messiness at the end of prefix-tree.