Ximin Luo

This issue is about getting something simple that is already part of the TLS 1.3 specs, into rustls for general use, outside of the context of any specific p2p protocol....

I tried screwing around with the config but it doesn't help either - `lconfig_hangingTypeSignature: true` doesn't affect this.

Debian Rust packaging is already done with a cargo wrapper tool that avoids network access, see https://salsa.debian.org/rust-team/debcargo-conf/ - rav1e can submit a PR for itself and any of its dependencies...

Unfortunately k, g, e, n are optimised out and printf() did not work for some reason, but the other variables are there, see further below. The test passes when you...

For completeness, `-O1` also makes the test fail.

The assembly is very heavily optimised and doesn't match the source code very well, things are run out-of-order. Here it is running in gdb's `layout asm`: ~~~~ 0x555555554da0 push %r15...

The test passes with `-O2` if you replace the platform-specific `count_leading_zeros` with the fallback definition in flint's longlong.h (outside of arb). i.e.: ```diff --- /usr/include/flint/longlong.h.orig 2017-10-21 01:38:05.824715526 +0200 +++ /usr/include/flint/longlong.h...

I'm starting to suspect it's an optimizer bug. I couldn't see anything obviously wrong with the code, though I'm not super-familiar with all the details of what is and is...

Right, I wasn't suggesting that you adopt the patch I added, it was just to demonstrate different test failures modes. With advice from #gcc I found a minimal optimiser setting...

Using [C-reduce](https://embed.cs.utah.edu/creduce/) I managed to get a [minimal example](https://paste.debian.net/992172/) that doesn't include any flint or arb code. I'm talking to various people to see if the C/asm code has any...