[Toybox] [PATCH] roadmap update: Android switched to toybox ls today.

[Rob Landley]

On 05/16/2015 03:43 PM, enh wrote:
> actually, this is the one outstanding patch you _shouldn't_ apply --- i
> had to revert the switch to ls late last night after it broke a
> surprising amount of infrastructure.

Yeah, I was kinda surprised. I applied your first "add selinux" patch
verbatim, but it builds with a warning under glibc. I have pending stuff
to fix in it, I posted my notes from my first look at it...

> (the toolbox ls didn't columnate,

I think I fixed that one here:

https://github.com/landley/toybox/commit/b18c7e8a59e61b8ec49e2d0f8cb2dda19b8f6a82

> and "adb shell ..." creates a pty on
> the remote side, so toybox ls sees that and defaults to -C rather than
> -1. callers can't use -1 to say what they want, because toolbox ls
> didn't support that, and even if we add it today they still need to deal
> with a mix of old and new devices.

In theory it should do the ANSI size probe to talk to your xterm. You
could also have "adb shell" export COLUMNS=1 to force -1 mode?

(I'm not sure what the desired behavior is here in "given x, toybox
should do y" terms. Is "adb shell" being used to run scripts? Provide an
interactive terminal? Dunno the use case.)

> similarly, the pty creation is on the
> adbd side, so although i'll fix that to be like ssh(1), that will only
> help new devices.

I'm confused: you get toybox when you upgrade to android m, what
infrastructure is _not_ upgraded when you do that, causing the version skew?

> i suspect we'll have to disable the isatty test in
> ls_main for the next ten years.

In theory isatty() can also be taught "if the terminal size is 0 by 0
when you query it, it's not real".

It's easy enough to make a kconfig switch so the behavior defaults to -1
instead of -C even when you have a terminal. (Or the test can be if
!CFG_TOYBOX_ANDROID...) I'd just like to understand th problem space a
little better first?

> there's also code that parses the
> toolbox ls -l format, which doesn't have the "nlink" column, so i'll
> need to find and fix those regular expressions. but first i need to talk
> to the ddmlib folks to see why they're parsing ls output at all, when
> using the adb sync protocol queries would seem like a better idea...)

We can add a flag to switch off that column, but given that posix is the
one that said to have that column:

http://pubs.opengroup.org/onlinepubs/9699919799/utilities/ls.html

That may not be the ideal solution...

> but there's plenty of other stuff like the hwclock fix and the find
> -inum addition and the xxd for pending to keep you busy,

I'm trying to catch up, but I get on a plane back to japan tomorrow
afternoon.

(Also, I'm supposed to port $DAYJOB's kernel to smp for sh2 (which is
fun because sh2 hasn't got an an atomic test-and-set so we may need to
change the vhdl again), and add dma to our drivers now the dma engine
can actually generate a completion interrupt, and finish forward porting
the the kernel to 4.0 (the ethernet driver uses obsolete infrastructure
to fake hardware timestamps and it broke for the third time between 3.19
and 4.0, I need to understand it so I can rip it out. Plus I spent half
the week unsuccessfully trying to make the h8 port's updated elf2flt
repo work with my last gplv2 release binutils 2.17, and I'm _still_
learning how bits of linking work...)

> plus -- as you
> said -- there's smack for the flight :-) (rather than make things more
> confusing and risk slowing things down even further, i'm waiting until
> the smack changes are in before i worry about SELinux. what i've seen
> looks plausible though so i'm not anticipating problems.)

There's a first pass review of that already posted, working on some of
that today...

> btw, on the subject of testing --- i noticed that a command that SEGVs
> gets a "PASS". i've been meaning to fix that and send a patch, but since
> it's been two weeks since i noticed already, i should at least report it...

Huh. It shouldn't. (Unless it was expected to produce no output? Output
to stderr can get ignored, and the return value's only meaningful if you
explicitly test for it ala "command && echo yes"...)

Where did you notice this? (Context?)

Rob