@NotsoanoNimus I started.

So far, std.hash.sha1.ll size reduced from 938620 bytes to 191464 bytes, where the original is 3.7% 181% faster.

938620 -> 238203 bytes
Original is 3.2% faster.

General rewrite; basically no difference, but is cheaper to compile and will compile smaller for less optimized builds.

I'm trying to make it identical to current master, and it seems to be a tiny difference in how the compiler handles the same algorithm inlined differently.

Nope. It's just the volatile erasure. Didn't realize not commenting it out in both parts was the only difference. Code is identical to master.

@lerno Do we omit state erasure on the grounds that this primitive is broken?

Maybe make secure erasure a compile-time option/$feature instead of removing it entirely?

Most SHA1 usage probably does not benefit from wiping, but some users may still want the hardening for HMAC/password-related contexts 🤷️

State erasure is good, but is it costly?

State erasure is good, but is it costly?

0.8% faster without

Ready to merge?

Something's wrong, the tests fail:

RROR: 'Assert "sha.final() == x"84983E44 1C3BD26E BAAE4AA1 F95129E5 E54670F1"" failed.'
  in std.os.posix.backtrace (/home/c3c/source/lib/std/os/posix/process.c3:91) [/tmp/tmp.n8MgTRHLfX/unit/testrun] [inline]
  in std.os.backtrace.capture_current (/home/c3c/source/lib/std/os/backtrace.c3:87) [/tmp/tmp.n8MgTRHLfX/unit/testrun] [inline]
  in std.core.builtin.print_backtrace (/home/c3c/source/lib/std/core/builtin.c3:227) [/tmp/tmp.n8MgTRHLfX/unit/testrun] [inline]
  in std.core.runtime.panic_test.20669 (/home/c3c/source/lib/std/core/runtime_test.c3:117) [/tmp/tmp.n8MgTRHLfX/unit/testrun] [inline]
  in std.hash.sha1_test.test_sha1_longer (/home/c3c/source/test/unit/stdlib/hash/sha1.c3:26) [/tmp/tmp.n8MgTRHLfX/unit/testrun] [inline]
  in std.core.runtime.run_tests (/home/c3c/source/lib/std/core/runtime_test.c3:333) [/tmp/tmp.n8MgTRHLfX/unit/testrun]
  in std.core.runtime.default_test_runner (/home/c3c/source/lib/std/core/runtime_test.c3:386) [/tmp/tmp.n8MgTRHLfX/unit/testrun] [inline]

Test failed ^^^ ( sha1.c3:26 ) Assert "sha.final() == x"84983E44 1C3BD26E BAAE4AA1 F95129E5 E54670F1"" failed.

Yikes, okay.

Silly me; handled. Improved my test: 0.63% faster without zeroing.

Interesting! This seems to have triggered a lowering bug. Don't change the code, I need to look at it.

Thank you!