[Toybox] macOS sitrep

[Rob Landley]

On 11/30/18 4:46 PM, enh via Toybox wrote:
> so, if you merge this afternoon's patches, you can build toybox for
> macOS. with two provisos:
> 
> 1. i'm only talking about a very limited .config, that's basically
> "the easy parts of the subset needed by the AOSP build". specifically:
> 
> CONFIG_TOYBOX_ICONV=y
> CONFIG_TOYBOX_UTMPX=y
> CONFIG_TOYBOX_FORK=y
> CONFIG_BASENAME=y
> CONFIG_CAT=y
> CONFIG_CHMOD=y
> CONFIG_CMP=y
> CONFIG_COMM=y
> CONFIG_CUT=y
> CONFIG_DIRNAME=y
> CONFIG_DU=y
> CONFIG_ECHO=y
> CONFIG_FALSE=y
> CONFIG_FILE=y
> CONFIG_GREP=y
> CONFIG_EGREP=y
> CONFIG_FGREP=y
> CONFIG_HEAD=y
> CONFIG_ID=y
> CONFIG_WHOAMI=y
> CONFIG_LN=y
> CONFIG_MKDIR=y
> CONFIG_OD=y
> CONFIG_PASTE=y
> CONFIG_PATCH=y
> CONFIG_PWD=y
> CONFIG_RM=y
> CONFIG_RMDIR=y
> CONFIG_SED=y
> CONFIG_SLEEP=y
> CONFIG_SLEEP_FLOAT=y
> CONFIG_SORT=y
> CONFIG_SORT_BIG=y
> CONFIG_SORT_FLOAT=y
> CONFIG_TAIL_SEEK=y
> CONFIG_TEE=y
> CONFIG_TRUE=y
> CONFIG_UNAME=y
> CONFIG_UNIQ=y
> CONFIG_WC=y
> CONFIG_XARGS=y
> CONFIG_DOS2UNIX=y
> CONFIG_UNIX2DOS=y
> CONFIG_HELP=y
> CONFIG_HELP_EXTRAS=y
> CONFIG_READLINK=y
> CONFIG_REALPATH=y
> CONFIG_SETSID=y
> CONFIG_TIMEOUT=y
> CONFIG_XXD=y
> CONFIG_HOSTNAME=y
> CONFIG_MD5SUM=y
> CONFIG_SHA1SUM=y
> CONFIG_MKTEMP=y
> CONFIG_TOYBOX=y
> CONFIG_TOYBOX_SUID=y
> CONFIG_TOYBOX_LSM_NONE=y
> CONFIG_TOYBOX_FLOAT=y
> CONFIG_TOYBOX_HELP=y
> CONFIG_TOYBOX_HELP_DASHDASH=y
> CONFIG_TOYBOX_I18N=y

Easiest thing might be to put that in a file and have a "make mac_defconfig"?

> i'm not sure what we want to do about that long-term. add `depends on
> TOYBOX_LINUX` to all the stuff like dmesg that's never going to work?
> (and probably to the "too much work for anyone to care" stuff like ps
> too.)

I might reopen the ps can of worms for bsd, or if darwin became an actual
runnable thing, but probably not for a proprietary OS. :)

> 2. lib/getmountlist.c doesn't compile. i have #ifndef __APPLE__ ...
> #endif around the whole file locally. macos seems to have a wholly
> different API, getmntinfo(3).

Thank you _ever_ so much, Posix.

> if/when i get around to porting toybox
> stat(1) to macOS i might have to fix this properly, but for now i'm
> punting. (df isn't used by AOSP and mount/umount seem to be in the
> "probably not worth it" territory mentioned in #1.)

I ranted about that in the posix section of roadmap.html. I tried to figure out
a proper way to do it and there isn't one. But there's a reason it _doesn't_
return struct mntent, and that's also why I broke commas.c out from that file
when it started getting more generically used.

It might be best to just move the remaining 3 mntent functions into portability.c...

> things like struct stat and struct statfs turn out to be quite
> different on macOS. but that's a problem for another day.
> ______________________________________________

In theory, stat has a posix subset. Dunno if dirtree adheres to it, though. (But
openat() is posix-2008!) And there's a posix statvfs too:

http://pubs.opengroup.org/onlinepubs/009604599/basedefs/sys/statvfs.h.html

I've tried not to be _gratuitously_ linux-specific. But there are large chunks
of things that just haven't got portable APIs. :(

Rob