android: Print error when trying to build on 32-bit host.

build/android/envsetup_functions.sh

Index: build/android/envsetup_functions.sh
diff --git a/build/android/envsetup_functions.sh b/build/android/envsetup_functions.sh
index 16ec004227aec6d8d547d3f810c8ecc3f2d4514b..b667abc0e0b45a16dec331eed4bbc642bbc82cbd 100755
--- a/build/android/envsetup_functions.sh
+++ b/build/android/envsetup_functions.sh
@@ -157,6 +157,7 @@ common_gyp_vars() {
 print_usage() {
   echo "usage: ${0##*/} [--target-arch=value] [--help]" >& 2
   echo "--target-arch=value     target CPU architecture (arm=default, x86)" >& 2
+  echo "--try-32                try building a 32-bit host architecture" >&2

[bulach, 2012/11/15 00:51:00]

to avoid potential confusion, maybe try-32-host?

[digit1, 2012/11/15 09:30:24]

That's actually a very nice suggestion, will do.

   echo "--help                  this help" >& 2
 }
 
@@ -167,11 +168,15 @@ print_usage() {
 # --help          Prints out help message.
 ################################################################################
 process_options() {
+  try_32bit_build=

[Isaac (away), 2012/11/15 01:09:16]

nit: this line not needed, there are very few commands in bash that
differentiate before a defined but blank variable and an undefined one.

[digit1, 2012/11/15 09:30:24]

That sounds like very bad coding style to me. You're assuming that the function
is never going to be called more than once, and that the variable is not already
defined.

That kind of thing is probably ok in a one-off script, but this gets sourced
into the user's shell. So if I followed your advice and did the following:

  . build/android/envsetup.sh --try-32
  . build/android/envsetup.sh

The try_32bit_build definition from the first line is sticky, and would force
the build to fail after that.

One more reason why sourced shell scripts are evil, imho :)

[Isaac (away), 2012/11/15 09:59:14]

And I would have gotten away with it, too, if not for this meddling source
syntax. 

Bash scripts are confusing.  Sourced bash scripts are evil. 

On 2012/11/15 09:30:24, digit1 wrote:

[felipeg, 2012/11/15 13:01:05]

Hey Isaac,
I agree that sourced bash scripts are evil, but we already do this. In fact it
is a common practice.
So the fact that it is evil doesn't change the need to make the fix as david is
doing.

I think we should always "declare" the shell variables before using.
And if needed, we might even export them.

On 2012/11/15 09:59:14, Isaac wrote:

   while [[ $1 ]]; do
     case "$1" in
       --target-arch=*)
         target_arch="$(echo "$1" | sed 's/^[^=]*=//')"
         ;;
+      --try-32)
+        try_32bit_build=true
+        ;;
       --help)
         print_usage
         return 1