if necessary, to help with transition, one could add a hidden flag `-log_when_issue_18315_occurred` that would log stacktrace (or maybe user defined function) when hitting this condition at runtime:
``` void main(){ fun(int.min); } void fun(int v){ writeln(v>0); } ``` dmd -log_when_issue_18315_occurred -run main.d WARNING_18315 at (shows stacktrace): old:true, new:false false On Sun, Feb 4, 2018 at 3:01 PM, Timothee Cour <thelastmamm...@gmail.com> wrote: > thanks @aG0aep6G for this PR https://github.com/dlang/dmd/pull/7841 to fix it. > this should be in point release because: > * ldc2 has correct behavior > * the bug disappears with `-O` > so the argument that ppl would rely on it is moot > > > On Fri, Feb 2, 2018 at 6:37 AM, Steven Schveighoffer via > Digitalmars-d-announce <digitalmars-d-announce@puremagic.com> wrote: >> On 2/2/18 4:29 AM, Seb wrote: >>> >>> On Friday, 2 February 2018 at 08:34:32 UTC, Antonio Corbi wrote: >> >> >>>> Wouldn't it be good to include a fix for errors like produced by int.min >>>> assigned to a variable >>>> (https://forum.dlang.org/post/p4l7kt$80d$1...@digitalmars.com) in a point >>>> release like this? >>> >>> >>> No, while I understand that you would like this to be fixed, this change >>> might be disruptive - you never know on what weird behavior people rely. >>> Anything potentially breaking existing code can't be part of a patch >>> release. >> >> >> In some cases, yes, we need to have a deprecation period as people may >> depend on the behavior. >> >> In this case, however, the codegen is simply wrong. It can be fixed >> immediately. I would hazard to guess that nobody is depending on int.min >> being greater than 0. >> >>> Also AFAICT no one has submitted a PR to fix the issue you referenced, so >>> it's a hypothetical question (for now). >> >> >> I would suggest to would-be fixers, just do the correct thing that may be >> less performant, and we can worry about optimizing later (and add a unit >> test of course!). There is nothing worse than a compiler that doesn't emit >> the code you expect it to. >> >> -Steve