[Ur] TechEmpower Benchmarks

[Adam Chlipala]

On 11/02/2013 05:55 PM, escalier at riseup.net wrote:
> I drafted a submission for the TechEmpower Web framework benchmarks
> (results here: http://www.techempower.com/benchmarks/ | specs here:
> http://www.techempower.com/benchmarks/#section=code&hw=i7&test=json).
>    

Oh, neato!  Around 2008 I started looking for existing benchmarks that 
would make for good comparisons with Ur/Web, and there were depressingly 
few out there.  It's good to learn about this one; and I'm very 
interested in helping get the Ur/Web tests into the official 
comparison.  (Though what I'd _really_ like would be a benchmark that 
measures programmer productivity somehow on less trivial tasks.)

> Would anyone with more Ur/Web expertise be willing to look over my
> submission?
> (https://github.com/pseudonom/FrameworkBenchmarks/blob/master/UrWeb/bench.ur)
>    

I winced reading the code to add "Date" and "Server" headers, but I see 
that these are a required feature. ;)

You can use the infix operator [%] instead of calling the [mod] function.

The first pattern below is redundant, I think, if you change the third 
pattern to just [_]:
>     val on = case qt of
>         None => Some 1
>       | Some ("queries", x) => read x
>       | Some _ => Some 1
...and I'm cross with the benchmark authors for the requirement to 
handle nasty query strings missing this parameter or giving it in the 
wrong format!

Maybe define a function to encapsulate all the boilerplate for returning 
a JSON chunk?  It could even include a call to [addHeaders].

What do you think about adding loud comments to this file, clearly 
delineating the different test types?

> Areas that I know need attention:
> - I defined my own `map2` rather than using `Top.map2` to avoid figuring
> out the typing for `Top.map2`
>    

[Top.map2] is for records, not lists, so it wouldn't have done what you 
want anyway.  I've added your function (generalized in the usual 
polymorphic way) to the standard library as [ListPair.map2].

After I added the function and looked more closely at where you're using 
it, I wonder why you don't just replace this code:
     rands' <- List.mapM (fn _ => rand) rq;
     let
       val rows' = map2 (fn x y => x -- #RandomNumber ++ {RandomNumber = 
clamp y}) rows rands'
with this code:
     rows' <- List.mapM (fn r => n <- rand; return (r -- #RandomNumber 
++ {RandomNumber = clamp n})) rows

> - To get a list of random Ints of length x, I `List.mapM (fn _ =>  rand)`
> over a range from 1 to x
>    

I believe you could do this more nicely, using standard library code 
that I _didn't_ just add (;]), with:
     List.tabulateM (fn _ => rand) x

> - Ur/Web seems to expect an 'int8' where an 'int' would suffice
> - Ur/Web seems to expect a 'text' where a 'varchar' would suffice
>    

Sure, but is this really a problem?  I view it more as avoiding 
unnecessary fanciness!  (E.g., having multiple integer types seems 
pretty retro to me, for most contexts.)