<rpm-devel@rpm5.org> is likelier the better place ...
On Jan 3, 2009, at 2:37 PM, Jeff Putsch wrote:
Howdy,Please pardon the cross-posting, I'm not sure which list is more appropriate for this post. If an openpkg-* list is more appropriate, please let me know.I've built RPM 5.1.6 with RPM_VENDOR_OPENPKG defined, expecting to be able to override the location of the platform file via the __platform macro at run-time.Unfortunately, neither setting __platform in macros files, nor setting it via "--define '__platform /path/to/platform'" on the command line causes RPM 5.1.6 to read the specified platform file.I've traced the "problem" behavior to the defaultMachine() function in lib/rpmrc.c.The first time defaultMachine() is called, 'cp = rpmExpand("%{? __platform}", NULL);' is executed, cp is ALWAYS NULL, causing the hardcoded path to the platform file to be used EVERY time.Subsequent calls to defaultMachine() are short-circuited by the gotDefaults static variable.Running under gdb to diagnose the problem reveals defaultMachine() is called 4 times prior to setDefaults() being called.If, in gdb, I set defaultMachine's "gotDefaults" static variable back to "0" AFTER setDefaults is called, then the 'cp = rpmExpand("% {?__platform}", NULL);' line expands properly and the __platform variable is set properly.The problems seems to occur because rpmReadConfigFiles() calls rpmRebuildTargetVars() BEFORE rpmReadRC() is called and rpmReadRC() seems to initialize the global macro space. I could be wrong about this, but that is what my new-to-RPM-eyes see.
There's a __HUGE__ amount (too much imho) of complexity boot-strapping RPM configuration.
Note that there is (usually) a %{_target} macro in the paths
to read macros from.
In order to do the expansion of %{_target}, then /etc/rpm/platform
needs to be read.
If %{__platform} instead of "/etc/rpm/platform" is used, then % {__platform} needs to be expanded.
But if %{__platform} is in a macro file, well, we start to come full circle.
What might save the mess is that %{__platform} is defined before % {__target} is needed.
But that forces initialization to be convolved with itself, some macro files can be read any time, other macro files have complicated pre-requisites that are data dependent.
I've attached a simple patch that is my current work-around to the problem.
If the patch is all that is needed (I'll leave to Ralf & OpenPKG to apply), then
that can probably be accomodated. But the chicken <-> egg problem is already starting to cause problems. hth tnx for the patch. 73 de Jeff
Thanks, Jeff. --- lib/rpmrc.c.orig 2008-10-12 03:38:51.000000000 -0700 +++ lib/rpmrc.c 2009-01-03 11:23:42.000000000 -0800 @@ -502,7 +502,7 @@ */ static void defaultMachine(/*...@out@*/ const char ** arch, /*...@out@*/ const char ** os)- /*...@globals rpmGlobalMacroContext, h_errno, fileSystem, internalState @*/ + /*...@globals rpmGlobalMacroContext, h_errno, fileSystem, internalState, defaultsInitialized @*/ /*...@modifies *arch, *os, rpmGlobalMacroContext, fileSystem, internalState @*/{ #if defined(RPM_VENDOR_OPENPKG) /* larger-utsname */ @@ -625,7 +625,7 @@ } if (gotDefaults) break; - gotDefaults = 1; + gotDefaults = gotDefaults && defaultsInitialized; break; } ______________________________________________________________________ RPM Package Manager http://rpm5.org Developer Communication List rpm-devel@rpm5.org
smime.p7s
Description: S/MIME cryptographic signature
