Non-SFI mode: Enable PnaclTransitional Non-SFI browser tests on x86-64 bots.

ppapi/ppapi_nacl.gyp

Index: ppapi/ppapi_nacl.gyp
diff --git a/ppapi/ppapi_nacl.gyp b/ppapi/ppapi_nacl.gyp
index bbad95528af93616ddf6cd8ee08409b312f8256a..9ff9798b05e7c50e13359909baf38bfc692c88f4 100644
--- a/ppapi/ppapi_nacl.gyp
+++ b/ppapi/ppapi_nacl.gyp
@@ -235,7 +235,7 @@
             },
           ],
         }],
-        ['disable_pnacl==0 and (target_arch=="ia32" or target_arch=="arm") and OS=="linux"', {
+        ['disable_pnacl==0 and (target_arch=="ia32" or target_arch=="x64" or target_arch=="arm") and OS=="linux"', {
           # In addition to above configuration, build x86-32 and arm nonsfi
           # .nexe files by translating from .pexe binary, for non-SFI mode PPAPI
           # testing.
@@ -244,7 +244,7 @@
             'nmf_nonsfi%': '<(PRODUCT_DIR)/>(nexe_target)_pnacl_nonsfi.nmf',
           },
           'conditions': [
-            ['target_arch=="ia32"', {
+            ['target_arch=="ia32" or target_arch=="x64"', {
               'variables': {
                 'enable_x86_32_nonsfi': 1,
               },
@@ -270,17 +270,26 @@
                 '--output=>(nmf_nonsfi)',
               ],
               'target_conditions': [
-                ['enable_x86_32_nonsfi==1', {
+                ['enable_x86_32_nonsfi==1 and "<(target_arch)"=="ia32"', {

[Mark Seaborn, 2015/01/17 00:50:47]

Can this be
  target_arch=="ia32"
rather than
  "<(target_arch)"=="ia32"
?

Or is there some Gyp oddity here?

[hidehiko, 2015/01/28 15:04:16]

target_arch is not defined at the moment. Pls see other similar examples in
Chrome's gyp.

+                  'inputs': ['>(out_pnacl_newlib_x86_32_nonsfi_nexe)'],
                   'action': [
                     '--program=>(out_pnacl_newlib_x86_32_nonsfi_nexe)',
                     '--arch=x86-32',
-                  ],
+                  ]
+                }],
+                ['enable_x86_32_nonsfi==1 and "<(target_arch)"=="x64"', {
+                  'inputs': ['>(out_pnacl_newlib_x86_32_nonsfi_nexe)'],
+                  'action': [
+                    '--program=>(out_pnacl_newlib_x86_32_nonsfi_nexe)',
+                    '--arch=x86-64',

[Mark Seaborn, 2015/01/17 00:50:47]

Shouldn't this be x86-32?

We shouldn't need to have an "x86-64(-nonsfi)" entry in the NMF that points to
"x86-32-nonsfi" binaries. :-)

I guess the right thing to do here is to tweak src/trusted/plugin/ to use
"x86-32-nonsfi" in x86-64 builds?  Or is that fiddly enough that we should leave
it to a later cleanup?

[hidehiko, 2015/01/28 15:04:16]

I thought, the arch in the program section of nmf is the chrome's (i.e. target)
arch, isn't it? Running x86-32-nonsfi on 64bit chrome sounds like an surprising
behavior, as we allows x86-64?

[Mark Seaborn, 2015/01/28 21:51:04]

The arch key of an NMF entry is supposed to correspond to the architecture of
the files that the NMF entry points to.

The developer shouldn't need to care about whether Chrome's renderer & browser
processes run as x86-32 or x86-64.  What matters is whether the nexe that Chrome
chooses is x86-32 or x86-64.
Why is that surprising?  Running x86-32-nonsfi on 64-bit Chrome is the plan.  As
your design doc says, we want to "support x86-32 BMM even when the rest of
Chromium is built as x86-64"
(https://docs.google.com/a/chromium.org/document/d/1Dn7lBloLlKq8JEIZYg7xzZhrXt...).

The only tricky thing here is that, in x86-64 Chromium builds, there is a
difference between:

 1) nacl_helper (old): uses "x86-64 non-sfi mode" (which the PNaCl toolchain
can't target)
 2) nacl_helper_nonsfi (new): uses "x86-32 non-sfi mode"

[hidehiko, 2015/01/29 14:43:12]

I see.

I look into it, but fallback logic applying only for nacl_helper_nonsfi x86-64
looks to need a bit complicated code.
So, how about keep the current "hack" for now? Considering;
1) Anyway, we need to clean up browser_tests on x64 Chrome (e.g. libcfree)
2) Implementing fallback logic unconditionally looks much simpler than applying
only nacl_helper_nonsfi.
3) We are sure that nonsfi on x64 is anyway testing only.

WDYT? Added TODO, instead.

+                  ]
                 }],
                 ['enable_arm_nonsfi==1', {
+                  'inputs': ['>(out_pnacl_newlib_arm_nonsfi_nexe)'],
                   'action': [
                     '--program=>(out_pnacl_newlib_arm_nonsfi_nexe)',
                     '--arch=arm',
-                  ],
+                  ]
                 }],
               ],
             },