Re:It is not a compiler error. It is never a compiler error ...
by whizzter on Monday February 24, 2025 @08:31AM
(#43159348)
Large functions is a particular breeding ground.
Ages ago working on PS2 games one of our guys had a particularly huge "do-animations-and-interpolations-and-state-and-everything-for-the-hero-in-one-huge-switch" thingy (not uncommon to encounter in games) that crashed the GCC, the function was split up.
In the sequel I think a similar function grew enough that not only had they the function but also split in multiple files to avoid miscompiles.
Most recently I was generating an ORM binding(C#) from the database model of an ERP system, for mysterious reasons the C# runtime was crashing without stacktraces,etc (no debugger help). Having seen things like this before I realized that one of the auto-generated functions was huge so I split it up in multiple units and lo-and-behold it worked.
(Having written a tiny JVM once I also remembered that jump instructions are limited to 64kb, not 100% if the .NET runtime inherited that... once it worked I didn't put any effort into investigating the causes).
Most of the time though compiler bugs aren't the worst (unless they help cause confusion in already hard scenarios).
Ages ago working on PS2 games one of our guys had a particularly huge "do-animations-and-interpolations-and-state-and-everything-for-the-hero-in-one-huge-switch" thingy (not uncommon to encounter in games) that crashed the GCC, the function was split up.
In the sequel I think a similar function grew enough that not only had they the function but also split in multiple files to avoid miscompiles.
Most recently I was generating an ORM binding(C#) from the database model of an ERP system, for mysterious reasons the C# runtime was crashing without stacktraces,etc (no debugger help). Having seen things like this before I realized that one of the auto-generated functions was huge so I split it up in multiple units and lo-and-behold it worked.
(Having written a tiny JVM once I also remembered that jump instructions are limited to 64kb, not 100% if the .NET runtime inherited that... once it worked I didn't put any effort into investigating the causes).
Most of the time though compiler bugs aren't the worst (unless they help cause confusion in already hard scenarios).