precompile: support @big_str parsed integers
#27144
Conversation
|
This is clever, but I don't feel it is sufficient. We should be able to handle non-literal bigints. Otherwise there are still surprises like |
|
In my experience, the issues in question have always been that literals are broken under pre-compilation. That pre-allocating constant objects doesn't necessarily quite work the same is well known and doesn't seem to come up often. |
vtjnash
changed the title
@big_str parsed numbers@big_str parsed integers
And switch `big` macro to throwing errors immediately (when we can give perfect line info), rather than deferring until until maybe at runtime
vtjnash
force-pushed
the
jn/precompile-bigint
branch
from
5833a9d
to
2c25cab
Compare
|
OK to merge? I'm hoping / expecting we'll modify GMP to solve this completely in the future (like I did recently for MPFR), but this'll at least handle a few cases for now |
|
@JeffBezanson? This seems better than nothing and it already works. |
|
I don't really think it's worth it --- it doesn't fix cases like #15722 or the use in stdlib/Random. |
|
Alright, lacking any helpful counter-proposal, I'll kill this for now |
In support of #26991 (enable
__precompile__flag by default), this ensures that parser-derived "big" numbers are restored completely at runtime, to remove that odd "gotcha".