On 01/21/2013 08:44 PM, Andrew Dunstan wrote: > > On 01/21/2013 04:32 AM, Craig Ringer wrote: >> On 01/01/2013 10:54 AM, Noah Misch wrote: >>> On Mon, Oct 15, 2012 at 07:53:51AM -0400, Noah Misch wrote: >>>> The only matter still requiring attention is a fix for >>>> IsoLocaleName(). >>> Following off-list coordination with Brar, I went about finishing up >>> this >>> patch. The above problem proved deeper than expected. For Windows >>> Vista, >>> Microsoft made RFC 4646 tags the preferred way to denote a locale in >>> Windows. >>> Microsoft calls them "locale names". Starting with Visual Studio 2012, >>> setlocale() accepts locale names in addition to all the things it >>> previously >>> accepted. One can now use things like "initdb --locale=zh-CN" and >>> "CREATE >>> DATABASE foo LC_CTYPE = 'pl'". This meant updating win32_langinfo() >>> and >>> find_matching_ts_config() to handle the new formats. In passing, I >>> fixed an >>> unchecked malloc() in win32_langinfo(). >>> >>> In addition to expanding the set of valid locale inputs, VS2012 >>> changes the >>> (undocumented) content of _locale_t to hold locale names where it >>> previously >>> held locale identifiers. I taught IsoLocaleName() to handle the new >>> material. >>> I also sought to improve the comments on IsoLocaleName(); its >>> significance was >>> not previously clear to me. This thread has some background: >>> http://archives.postgresql.org/message-id/4964b45e.5080...@hagander.net >>> >>> Though I'm not entirely sanguine about digging into the >>> officially-opaque >>> _locale_t, we have been doing it that way for several years. None >>> of the >>> alternatives are clearly-satisfying. In particular, I found no good >>> way to >>> look up the code page corresponding to a locale name on pre-Vista >>> systems. >>> The CRT itself handles this by translating locale names to locale >>> identifiers >>> using a lookup table. The Gnulib "localename" and "setlocale" >>> modules are >>> also interesting studies on the topic. >>> >>> In previous reviews, I missed the need to update pgwin32_putenv(). The >>> addition of VS2010 support had also missed it, so this catches up. >>> That >>> function has other problems, but I will hold them for another patch. >>> >>> Tester warning: if you currently have some form of VS2010 installed, >>> including >>> the compilers of Windows SDK 7.1, beware of this problem: >>> http://stackoverflow.com/questions/10888391/link-fatal-error-lnk1123-failure-during-conversion-to-coff-file-invalid-or-c >>> >> FYI, you can properly fix this without uninstalling anything, giving >> you a system with a working VS 2012 as well as working SDK 7.1 / VS >> 2010 SP1 32-bit and 64-bit compilers. >> >> Install the tools in the following order: >> >> * VS Express 2010: >> http://www.microsoft.com/visualstudio/eng/products/visual-studio-2010-express >> * Windows SDK 7.1 >> * VS 2010 SP1: >> http://www.microsoft.com/en-au/download/details.aspx?id=23691 >> * VS 2010 SP1 Compiler Update: >> http://www.microsoft.com/en-au/download/details.aspx?id=4422 >> * VS Express 2012 >> >> Note that SDK 7.1 and VS 2010 will fail to install if you have a >> newer version of the Visual c++ 2010 runtime installed. Newer >> programs often install this for you. As a workaround, you must >> uninstall the newer runtime, install Visual Studio, the SDK, and >> service pack, then reinstall the newer runtime to get the programs >> that require it to work again. >> >> I've written more about both of these in the TROUBLESHOOTING section >> of https://github.com/2ndQuadrant/pg_build_win/blob/master/README.md >> > > That's useful information, which should perhaps be put somewhere more > obvious for people to find, like the wiki. > > I realise you're doing a lot of stuff, but you seem to be ahead of > just about everyone (including me and I suspect Magnus) on this, so > maybe you could take a peek at this patch?
It's building on my Jenkins instance at the moment, to make sure it doesn't break VS 2010 / Windows SDK 7.1 . I've merged it into HEAD and pushed to https://github.com/ringerc/postgres/tree/vs2012; the only conflict was a trivial one in the docs. -- Craig Ringer http://www.2ndQuadrant.com/ PostgreSQL Development, 24x7 Support, Training & Services
