Avoid too long command lines.

build/java.gypi

Index: build/java.gypi
diff --git a/build/java.gypi b/build/java.gypi
index ec8431b92da2e05417a7e716c1eb0d9ad2b69435..444cad439971c3e7e14a1e72bb484bbb428a938a 100644
--- a/build/java.gypi
+++ b/build/java.gypi
@@ -153,6 +153,10 @@
             # resources in dependencies can be resolved.
             'all_res_dirs': ['<@(res_input_dirs)',
                              '>@(dependencies_res_input_dirs)',],
+            # Write the inputs list to a file, so that the action command
+            # line won't exceed the OS limits when calculating the checksum
+            # of the list.
+            'inputs_list_file': '<|(inputs_list.tmp <@(_inputs))'

[cjhopman, 2013/07/25 15:35:17]

I think this file is written into the source tree, does .gitignore need any
entries for it? I didn't see any in a quick look for the usage of this gyp thing
in WebKit so maybe its not needed.

[cjhopman, 2013/07/25 15:35:17]

Since inputs contains things that are late expanded (use > instead of <), both
the expansions here probably need to be late.

[newt (away), 2013/07/25 17:55:13]

Could we put this in the out/ folder? in
<(intermediate_dir)/process_resources_inputs or something similar.

[cjhopman, 2013/07/25 18:05:26]

You can't write to the output directory at gyp time (it's not defined until
build time). I don't know if you even have access to some root of the output
directories.

[newt (away), 2013/07/25 18:11:40]

what happens if you try to write to the output directory at gyp time?
<intermediate_dir) has a valid value...

in any case, I imagine we need to write to separate files to keep parallel gyp
from exploding.

[cjhopman, 2013/07/25 18:17:50]

What happens is generator dependent, on ninja it expands to something like:
$!PRODUCT_DIR/chrome_java

But, this makes me notice that the filename here should definitely incorporate
the target_name otherwise multiple java targets in the same gyp file (or with
gyp files in the same directory) will be writing to the same inputs list file.

[whywhat, 2013/07/25 20:01:48]

.tmp files were in the untracked section of my git status.
Added a .gitignore entry for *.tmp

[whywhat, 2013/07/25 20:01:48]

Done.

[whywhat, 2013/07/25 20:01:48]

Done.

[whywhat, 2013/07/25 20:01:48]

Done.

[whywhat, 2013/07/25 20:01:48]

Ack

On 2013/07/25 18:05:26, cjhopman wrote:

[whywhat, 2013/07/25 20:01:48]

Ack

On 2013/07/25 17:55:13, newt wrote:

           },
           'inputs': [
             '<(DEPTH)/build/android/gyp/util/build_utils.py',
@@ -179,7 +183,7 @@
             # Add hash of inputs to the command line, so if inputs change
             # (e.g. if a resource if removed), the command will be re-run.
             # TODO(newt): remove this once crbug.com/177552 is fixed in ninja.
-            '--ignore=>!(echo \'>(_inputs)\' | md5sum)',
+            '--ignore=>!(cat <(inputs_list_file) | md5sum)',

[cjhopman, 2013/07/25 15:35:17]

inputs_list_file might need to be late expanded like above.

Also, md5sum can take the file as its argument instead of cat-ing the file and
piping it.

[whywhat, 2013/07/25 20:01:48]

Done.

           ],
         },
         # Generate API 14 resources.