Re: [jamboost] M6: Memory Consumption

From: Vladimir Prus (ghost_at_[hidden])
Date: 2003-07-25 08:13:51

Hi Jürgen,

> Ok. First, I observe the same behaviour with bjam -n.
> It uses 230 MB for header-scannning, I think and then starts using up
> memory while putting out 20x "patience".

> I've done a quick check with M5 and it works fine. It needs 228 MB for
> the first step (header-scannnig) and then about 260-280 MB.
> The interesting thing is that M5 reports
>
> bjam Version 2
> ...patience...
> 4x patience
> ...patience...
> ...found 25611 targets...
> ...updating 1 target...
>
> and M6 (current CVS) reports:
> bjam Version 2
> ...patience...
> 22 x patience
> ...patience...
> ...found 39385 targets...
> ...updating 1 target...
>
> So, M6 even somehow "finds" more targets then M5. Any Ideas ?

I'd say this is not surprising. I've changed the way header dependencies are 
handled (actually grabbed from Perforce Jam) and it adds some internal nodes. 
But... that happened for M5.

Could you please try getting CVS state as of 2003-07-10 and compare it's 
performance with M5 (again using -n switch).

TIA,
Volodya