Hi Antony, My fixes arrived at 33433. Did you sync that before trying the above? I hadn't validated command line builds (I'm checking that myself now).
Oh by the way, I assume you're using /Out because of devenv.exe 's weird console piping behavior. Something we learned with the bots is that if you use devenv.com, normal piping works correctly, ie you can use |less. So there were 2 core problems that I've patched up, but which need to be done properly, and a 3rd which was fixed about a week ago: 1. For ~1-2 months now, there has been an issue in which gyp generates different vcproj files (some slashes flipped the wrong way) when run under cygwin (this got broke at some point). This does not actually affect all cygwin users, because of the behavior of depot_tools. Historically depot_tools installed its own python, which took precedence over cygwin python. This meant that if you used gclient runhooks, you'd always be using win32 python. At some point in the last 3 months or so, depot_tools was changed, so that it only installs its own python if one is not already in the path. So cygwin users who have cygwin python installed at the time they first installed depot_tools, will end up using gyp with cygwin python. These cygwin generated vcproj files fail for several modules, notably nacl. I've corrected this in a somewhat hacky way, but the two now are basically identical. 2. Nacl currently makes extensive use of msvs_cygwin_shell: 0. This option in gyp was intended to allow direct access to cmd on windows. Most targets in chrome with actions/rules do not use this option. This option was intended as a backdoor for things like running setup_mount to get the hermetic cygwin initialized. Unfortunately, nacl's scons build made extensive use of scons's more flexible command line handling. In transitioning to gyp, several command lines where handled by using conditionals on platform, and using platform specific piping etc. msvs_cygwin_shell: 0, makes no attempt to setup cygwin or python in the path. As a result, at some point nacl folks added PYTHON_EXE, wrapper scripts, and other gyp variables to 'fix' the problem. These scripts didn't take into account the possibility that devenv.exe could be invoke with cygwin stuff in the path, and so fail when devenv is started from the command line. I've corrected these scripts so that they work in the presence of cygwin. This is, however, a stopgap. What really needs to happen is that all actions/rules in nacl's gyp files should NOT use msvs_cygwin_shell. This will mean adding some additional python wrapper scripts / changing the assumption in existing scripts that the build system is capable of handling piping. 3. A little while back we had a miscommunication about output directories. Nacl has started generating 32 and 64 bit output. A change went into nacl's common.gypi which causes output to go to Debug-Win32 / Debug-x64. Since none of chrome's infrastructure knows to clobber these directories, incremental builds can hide problems. This got fixed about a week ago. So all the output should be under Debug. So far so good on the command line build I started at the beginning of this email. I'll drop by and see if you're still hitting issues this afternoon. -BradN On Tue, Dec 1, 2009 at 9:39 AM, Brad Chen <bradc...@google.com> wrote: > [+ilewis] > > I'm happy to hear the Visual Studio GUI build is working; I think this was > the fix Brad Nelson submitted yesterday. I don't think we've tested a build > using a script such as you suggest. We're checking it out now. > > I apologize for the disruption. The diversity of ways people build on > Windows and Linux is challenging, and we have difficulty supporting kinds of > builds we don't know about. The Chrome Windows build looks greenish. It's > much easier for my team to support Windows/Linux build environments if they > are represented in the buildbots. If somebody has ideas for how this > specific issue can be better represented in the Chrome buildbots I'm very > interested. > > Brad > > On Tue, Dec 1, 2009 at 9:24 AM, Antony Sargent <asarg...@chromium.org>wrote: > >> I've tried a few combinations of deleting my Debug directory and running >> "gclient runhooks", and I *think* I've narrowed the problem down to the >> build failing if do a command-line build, but working ok if I build from >> within the Visual Studio GUI. The script I have which does the (now failing) >> command line build does: >> >> devenv.exe chrome.sln /Build Debug /Out C:\\tmp\\build.log /Project chrome >> >> This worked fine until quite recently. >> >> >> On Tue, Dec 1, 2009 at 12:18 AM, Peter Kasting <pkast...@google.com>wrote: >> >>> On Mon, Nov 30, 2009 at 10:01 PM, Antony Sargent >>> <asarg...@chromium.org>wrote: >>> >>>> I just updated to revision 33425, and a clean build (manually deleted >>>> Debug directory before opening .sln file) gives the following error in >>>> service_runtime_x86. I'm running Visual Studio 2008 on Vista x64. Anyone >>>> else seeing this, or know what might be the problem? >>> >>> >>> NaCl has had problems building with Cygwin for almost two weeks. I >>> believe Brad Nelson just fixed things, so try wiping out your Debug/ dir, >>> re-syncing, and rebuilding. (I haven't yet tested this myself.) >>> >>> PK >>> >>> -- >>> Chromium Developers mailing list: chromium-dev@googlegroups.com >>> View archives, change email options, or unsubscribe: >>> http://groups.google.com/group/chromium-dev >> >> >> -- >> Chromium Developers mailing list: chromium-dev@googlegroups.com >> View archives, change email options, or unsubscribe: >> http://groups.google.com/group/chromium-dev >> > > -- Chromium Developers mailing list: chromium-dev@googlegroups.com View archives, change email options, or unsubscribe: http://groups.google.com/group/chromium-dev
