[totetsu]

What kind of assumptions would badly-written applications make to make themselves impractical to cross-build?

[bregma]

Othar than word size (eg. i386 vs. x86_64), endianness (eg. PPC vs. x86), alignment (eg. ARM vs. x86) there's plenty of other low-level issues like parameter passing conventions, call frame handling, exception handling, presence or absence of optional instructions or registers, and a number of other subtle gotchas that could affect a port. Instruction probing is one of the ones we run into where everything builds find and fails at runtime.

If you're working at a high enough level you're probably fine. Like the difference between being an auto mechanic and using the app to hail a ride.

[smallpipe]

You're missing the biggest one, concurrency model.

[capitol_]

In my packaging work I encounter hard coded assumptions about pointer width, endianness and presence of atomic operations.

[sanxiyn]

All Debian packages are built natively. There is no cross build going on here.

[pabs3]

During the initial port bootstrap process, the initial package set is cross-built using rebootstrap:

https://wiki.debian.org/PortsDocs/New#Bootstrap

Perhaps in the future Debian will be able to have cross-build-only architectures, that have hardware too slow to build software.

[ilyt]

Assembler optimizations without fallback to pure C for other architectures for example.