build.rs: Add feature 'lowmemory' to reduce memory usage

[laanwj]

Currently, this only set ECMULT_WINDOW_SIZE to 4 instead of 15.

Fixes #139.

[elichai]

ACK 09037fc

[apoelstra]

If you add a commit bumping the version to 0.15.1 (and setting the date in CHANGELOG.md) I can merge and publish today

[laanwj]

Whoops, I didn't realize that the version is already at 0.15.1, https://github.com/rust-bitcoin/rust-secp256k1/blob/master/Cargo.toml#L4
So I think I should bump it to 0.15.2 and change the CHANGELOG.md entry to that?

[real-or-random]

Do we think 3 is good is that "too low" to be reasonable?

[apoelstra]

Oh, yeah, 0.15.1 just had a docs.rs fix.

@real-or-random for embedded applications where speed is nonessential I think we want "as low as possible".

[laanwj]

@real-or-random for embedded applications where speed is nonessential I think we want "as low as possible".

3 seems to be fine for one-shot purposes such as verifying firmware signatures. The speed-up is mostly interesting in applications that verify large numbers of signatures.

I once did some benchmarks here: bitcoin-core/secp256k1#614 (comment)

So from 3 to 15 is at most a 18% speed-up, where going to 6 already accounts for about half of that. Even with the lowest setting I'd say it's fast enough for boot-loader use of verifying one or a few signatures.

[laanwj]

added version bump to 0.15.2 and updated the changelog accordingly !

[real-or-random]

@real-or-random for embedded applications where speed is nonessential I think we want "as low as possible".

If you want as low as possible, that's "2" and not "3".

@laanwj Can you update this?

(I restarted Travis, which had some hiccup).

[apoelstra]

Travis is still having problems :/

[laanwj]

If you want as low as possible, that's "2" and not "3".
@laanwj Can you update this?

That would be 340 to 170 bytes? don't know if that's worth it. secp256k1 itself is 70k+ compiled.

[real-or-random]

I'm not sure either; that's why I started the discussion about a reasonable low value. I think 3 is good, I 'd even pick 4.

[real-or-random]

If we want anything > 2, then we shouldn't claim that this is the lowest accepted value.

[laanwj]

OK, I'll set it to 4 and remove that comment.

[apoelstra]

utACK

[apoelstra]

Published 0.15.2