Re: [Gnash-commit] gnash libbase/tu_swap.h libamf/amf.cpp server/p...

[Martin Guy]

> why drop compile time tests ?

Because it trades a negligible speed thing against eternal porting
maintenance. It costs one assignment and a switch or a test (I didn't
figure out how to statically initialise one member of a union in C++)

John said they had used this technique in another project with great
relief. It went against the grain for me, but thinking on it I agreed.
Then I realised it would also let us handle hosts with middle-endian
IEEE754 doubles, which are still in use on ARM systems (and are a
Flash movie's native double format!), as well as opening a path for
non-IEEE754 floating point systems like IBM and Vax by having a final
catch-all option to interpret the floating point words explicitly.

> The boost endian.hpp header has the correct macros on all
> the platforms we have to support, so we should use them for endianess
> detection.

They don't seem to detect middle-endian (PDP-endian) doubles, which
still exist today on ARM systems that do floating point emulation
using the nwpfe emulator that is in the linux kernel (the real
PDP-endian "FPA" that it emulates stopped being made in silicon in the
90s, but its ghost will not rest).

The point of compile-time tests is speed and code size, but there's
virtually nothing to be gained here. However, if someone wants to put
compile-time tests back in where integers are involved, I have no
problem with that:
$ find . -type f | xargs -d\\n grep -i endian
should find them all.

    M

while (chit_chat_over_trivia) actual_work--;