Re: [Sbcl-devel] Slow clsql compilation

[Nikodemus S. <nik...@ra...>]

Christophe Rhodes <cs...@ca...> writes:

> The time seems to be spent in this function, which has an optimize
> declaration of (debug 3) (speed 1); (debug 3) in SBCL means munging
> the code substantially, making it very large -- and the function is
> large to begin with -- and there are algorithms in the compiler which
> are asymptotically slower than O(n).
>
> For what it's worth, the compiler seems to spend its time in
> SB-C::PROPAGATE-LIVE-TNS called from SB-C::LIFETIME-FLOW-ANALYSIS.  I
> don't know if this has changed recently, but I don't think it has.

In the sense that earlier (DEBUG 3) killed the ability of the compiler
to reason about the code almost totally.

Now it can make a better job of high-debug code, but it can obviously
still take uncomfortably long. 

There are some things that we could still do: for example making 
(> debug speed) inhibit inline-expansion might help with this.

Cheers,

  -- Nikodemus              Schemer: "Buddha is small, clean, and serious."
                   Lispnik: "Buddha is big, has hairy armpits, and laughs."