On Friday, 22 June 2018 at 22:25:35 UTC, Steven Schveighoffer
wrote:
I've gone through the "annotate this and continue" exercise,
and it is not fun. Especially when the template is used in many
places, you have to make a copy of the template to play with.
Aye. I think that's the most straightforward implementation
though: make the compiler on the inside just set the inferred
flag, then rerun the semantic and collect the errors until it
gets to the bottom.
(then perhaps cut out some of the errors from the middle for
brevity of the message, just showing the top and bottom parts of
the call chain, since that is most likely what the user actually
would want to know to fix it)