tree 365a0651fe39f142f9fbf895876887b37bea4491
parent 010da0b2f48dc92be2f79495fd4551c92351868f
author Ivan Levkivskyi <levkivskyi@gmail.com> 1693300672 +0100
committer GitHub <noreply@github.com> 1693300672 +0100
gpgsig -----BEGIN PGP SIGNATURE-----
 
 wsBcBAABCAAQBQJk7bfACRBK7hj4Ov3rIwAApS4IAKaDg93FFOE4FBonwbnScowp
 s8xhSDLUf6eCUVRFhamih1d03xJJCm3kaNJueeVoMChllQV0S9nptXmoie1JvEkm
 CsCgHla2FUv3361jxolxk/f94k2BAOKOV/ugd/jevjYbwWJxi++P2Egf7fS0sHw5
 VpxGH+57Y8A9Hi+0qWq9gWXtxwqO0TMX2AOP+H0W/ytSZ+Fe+QrHiSE3lHhPQ4c5
 FT3+Fg+AeR1PaCX6V95oc4yJJlZl+Irx5kemKrtji56HQAt05wQKRKfK8kkvWWe5
 uMpPmsI597V2E/jEfDl8UFn0Xau8RlsG6LbeF0p5pRCo9bWkhpdJNCKS2qA8p+4=
 =xatw
 -----END PGP SIGNATURE-----
 

Optimize Unpack for failures (#15967)

This is a small but possibly important PR. Wherever possible we should
represent user error and/or failed type inference as `*tuple[Any,
...]`/`*tuple[<nothing>, ...]`, rather than
`Unpack[Any]`/`Unpack[<nothing>]` or plain `Any`/`<nothing>`. This way
we will not need any special casing for failure conditions in various
places without risking a crash instead of a graceful failure (error
message).