[ale] Yea!! Android emulator working after 1/2 hr hair pulling

[Michael B. Trausch]

On 10/14/2013 10:21 AM, Ron Frazier (ALE) wrote:
> They REALLY should put an appropriate descriptive error message.

It's like anything else---once you know what it means, it makes sense. 
The world is full of cryptic information that we humans have to decode. 
The POSIX interface, however, is well-defined and there are a finite
number of things that can cause an error.

> Like, "The superapp application failed to run because its executable bit is off."  or "The emulator application failed while trying to load xxx.yyy library."

POSIX errors are sufficiently descriptive.  Consider what happens when
you're sitting at a shell prompt and run "ls":

 1. The shell will look for a file named "ls" to execute.  This involves
    searching the PATH for a file that is named "ls" and is executable.
 2. If a file named "ls" is not found in the PATH, the shell says
    "command not found".
 3. If a file named "ls" is in the PATH, but it is not executable, the
    shell says "command not found".
 4. If a file named "ls" is found and it is executable, the shell asks
    the kernel to run it.

Now, what happens if "ls" is a binary, is executable, and is of the
wrong architecture?

For that information, refer to the execve(3) man page (say "*man 3
execve*" to the shell).

I'll continue, though, about how program execution works here.

Let's say that you have ls, and it is for ARM.  When you try to run it:

[mbt at aloe ~]$ file ls
ls: ELF 32-bit LSB executable, ARM, EABI5 version 1 (SYSV), dynamically
linked (uses shared libs), for GNU/Linux 2.6.32,
BuildID[sha1]=be286f4b922f48718ca564b4eb9eb406407e4f3d, stripped
[mbt at aloe ~]$ ./ls
/lib/ld-linux.so.3: No such file or directory

Why? I don't have the file!  The kernel went ahead and attempted to load
the ELF interpreter mentioned in the file, because there is a chance
that the ELF interpreter might know how to read the file and give the
kernel guidance on how to run it.  (Yes, ELF's format does actually
support multiple-purpose usage; it doesn't just have to be a conatiner
for native code!)  Additionally, it is important to note that on my
system (Fedora 20 prerelease), an ARM ELF image will be executed with
/usr/bin/qemu-arm automatically.

No such file or directory makes sense here, because /it is the error
which caused the execution to fail./  What happens if I have that file,
then?

To find out, I've symlinked my dynamic loader to the requested name,
ld-linux.so.3.  Now:

[mbt at aloe ~]$ ./ls
/lib/ld-linux.so.3: Invalid ELF image for this architecture

Hrm.  Why?

Because the ARM code emulator doesn't know how to talk to ELF x86-64
libraries!

What happens if I remove its ability to load the ARM emulator, though? 
To do that, I will remove the /proc/sys/fs/binfmt_misc/qemu-arm file
from my system (it will come back at next boot):

echo -1 > /proc/sys/fs/binfmt_misc/qemu-arm

This removes the entry from the system.  Now what happens?

[mbt at aloe ~]$ ./ls
-bash: ./ls: cannot execute binary file

This is what happens when you have something that truly cannot be
executed.  To find out the why you can use strace:

[mbt at aloe ~]$ strace ./ls
execve("./ls", ["./ls"], [/* 73 vars */]) = -1 ENOEXEC (Exec format error)
write(2, "strace: exec: Exec format error\n", 32strace: exec: Exec
format error
) = 32
exit_group(1)                           = ?
+++ exited with 1 +++
[mbt at aloe ~]$

> (Yes, to any newbies reading, you may get a file not found error on Ubuntu 12.04 and Mint 13 and possibly others if the executable bit in the file properties is off.

Of course you will---the executable bit isn't set.  It's like having
CMD.EX_ instead of CMD.EXE on a Windows box; it won't be found it if
doesn't comply with the expectations that the system has of it.

> Well horsefeathers.  I just tested this by taking away the executable permission from the LS command.  It then gives a permissions error.  However, I'm pretty sure I've seen this manifest as a file not found error too.)

No, you should have an error like this:

[root at aloe ~]# ls
-bash: /bin/ls: Permission denied

Which will happen if ls has no permission for execute.

If a directory in your PATH has no execute permissions and you're using
an older system, you'll get a more generic permission denied message
that doesn't actually mention the file itself.  I can't replicate that
on my system because it doesn't follow that behavior anymore; bash (or
exec(), not sure which) is actually skipping any directories in $PATH
with bad permissions now.  Yay!


> Or better yet, (what I said above) + "error code 2342".  Which you could then google.  And / or, print a trace of the last 10 commands.

POSIX errors already have codes, and they're much easier to remember. 
They are symbols; e.g., E2BIG, EACCES, EINVAL, ELOOP.

However, we don't expose them to the user always, because we don't need
to.  If you are asking your shell to execute something and it cannot do
so, it is up to you to investigate why.  This is standard practice.

> What really confused me is that I'd already run the emulator, albeit insanely slowly, on my laptop.  This missing file phenomena never happened.  But ... that vm is 32 bit, not 64 bit.

That'd be why.

> These events of pounding on something that is already (as far as I could tell) properly configured put serious dents in my already dented sanity, at least temporarily.

You just have gaps in your knowledge, that's all.  You need to spend
some time studying the system that you're working with */before/* trying
to do very advanced things with it.

You should not be developing for any system if you cannot troubleshoot a
file execution issue relatively quickly.  You do need to know the system
you're developing /*on*/, or you're going to have a very hard time
proceeding.

    --- Mike

-- 
Naunet Corporation Logo 	Michael B. Trausch

President, *Naunet Corporation*
? (678) 287-0693 x130 or (855) NAUNET-1 x130
FAX: (678) 783-7843

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.ale.org/pipermail/ale/attachments/20131014/b2f42c33/attachment-0001.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: gihibjii.png
Type: image/png
Size: 1701 bytes
Desc: not available
URL: <http://mail.ale.org/pipermail/ale/attachments/20131014/b2f42c33/attachment-0001.png>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 901 bytes
Desc: OpenPGP digital signature
URL: <http://mail.ale.org/pipermail/ale/attachments/20131014/b2f42c33/attachment-0001.sig>