diff options
-rw-r--r-- | bpkg/system-package-manager-fedora.cxx | 107 |
1 files changed, 87 insertions, 20 deletions
diff --git a/bpkg/system-package-manager-fedora.cxx b/bpkg/system-package-manager-fedora.cxx index 8ea6779..3b79c50 100644 --- a/bpkg/system-package-manager-fedora.cxx +++ b/bpkg/system-package-manager-fedora.cxx @@ -2651,7 +2651,7 @@ namespace bpkg dir_path licensedir; dir_path build2dir; - // Note that the ~/rpmbuild/{BUILD,BUILDROOT,RPMS,SOURCES,SPECS,SRPMS} + // Note that the ~/rpmbuild/{.,BUILD,BUILDROOT,RPMS,SOURCES,SPECS,SRPMS} // directory paths used by rpmbuild are actually defined as the // %{_topdir}, %{_builddir}, %{_buildrootdir}, %{_rpmdir}, %{_sourcedir}, // %{_specdir}, and %{_srcrpmdir} RPM macros. These macros can potentially @@ -2691,6 +2691,15 @@ namespace bpkg expressions.push_back ("%{_rpmdir}/%{_rpmfilename}"); + // Note that we rely on the fact that these macros are defined while + // refer to them in the spec file, etc. Thus, let's verify that and fail + // early if that's not the case for whatever reason. + // + expressions.push_back ("%{?_rpmdir}"); + expressions.push_back ("%{?_rpmfilename}"); + expressions.push_back ("%{?_usrsrc}"); + expressions.push_back ("%{?buildroot}"); + // Note that if the architecture passed with the --target option is // invalid, then rpmbuild will fail with some ugly diagnostics since // %{_arch} macro stays unexpanded in some commands executed by @@ -2725,13 +2734,13 @@ namespace bpkg return r; }; - auto pop_dir = [&expansions, &expressions] () + auto pop_path = [&expansions, &expressions] () { assert (!expansions.empty ()); try { - dir_path r (move (expansions.back ())); + path r (move (expansions.back ())); if (r.empty ()) fail << "macro '" << expressions.back () << "' expands into empty " @@ -2748,6 +2757,11 @@ namespace bpkg } }; + auto pop_dir = [&pop_path] () + { + return path_cast<dir_path> (pop_path ()); + }; + // The source of a potentially invalid architecture is likely to be the // --architecture option specified by the user. But can probably also be // some mis-configuration. @@ -2755,7 +2769,13 @@ namespace bpkg if (expansions.back ().empty ()) // %{?_arch} fail << "unknown target architecture '" << arch << "'"; - pop_string (); // We only need %{?_arch} expansion for the verification. + // We only need the following macro expansions for the verification. + // + pop_string (); // %{?_arch} + pop_dir (); // %{?buildroot} + pop_dir (); // %{?_usrsrc} + pop_string (); // %{?_rpmfilename} + pop_dir (); // %{?_rpmdir} rpmfile = pop_string (); specdir = pop_dir (); @@ -3135,6 +3155,10 @@ namespace bpkg // and recommendations, etc) with the view that this can be used as a // starting point for manual packaging. // + // NOTE: if changing anything here make sure that all the macros expanded + // in the spec file unconditionally are defined (see above how we do + // that for the _usrsrc macro as an example). + // try { ofdstream os (spec); @@ -3569,34 +3593,77 @@ namespace bpkg { os << "%undefine _debugsource_packages" << '\n'; - // Append the -ffile-prefix-map option (if specified) which is used - // to strip source file path prefix in debug information (besides - // other places). By default it is not used, presumably since we - // disable generating the -debugsource sub-packages. We change it to - // point to the bpkg configuration directory. Note that this won't - // work for external packages with source out of configuration - // (e.g., managed by bdep). + // Append the -ffile-prefix-map option which is normally used to + // strip source file path prefix in debug information (besides other + // places). By default it is not used since rpmbuild replaces the + // recognized prefixes by using the debugedit program (see below for + // details) and we cannot rely on that since in our case the prefix + // (bpkg configuration directory) is not recognized. We need to + // replace the bpkg configuration directory prefix in the source + // file paths with the destination directory where the -debugsource + // sub-package files would be installed (if we were to generate it). + // For example: + // + // /usr/src/debug/foo-1.0.0-1.fc35.x86_64 // - // @@ Supposedly this code won't be necessary when we add support for - // -debugsource sub-packages. + // There is just one complication: // - // Note that adding this option may also result in notification - // messages for probably all the source files as following: + // While the generation of the -debugsource sub-packages is + // currently disabled, the executed by rpmbuild find-debuginfo + // script still performs some preparations for them. It runs the + // debugedit program, which, in particular, reads the source file + // paths from the debug information in package binaries and saves + // those which start with the ~/rpmbuild/BUILD/foo-1.0.0 or + // /usr/src/debug/foo-1.0.0-1.fc35.x86_64 directory prefix into the + // ~/rpmbuild/BUILD/foo-1.0.0/debugsources.list file, stripping the + // prefixes. It also saves all the relative source file paths as is + // (presumably assuming they are sub-entries of the + // ~/rpmbuild/BUILD/foo-1.0.0 directory where the package archive + // would normally be extracted). Afterwards, the content of the + // debugsources.list file is piped as an input to the cpio program + // executed in the ~/rpmbuild/BUILD/foo-1.0.0 directory as its + // current working directory, which tries to copy these source files + // to the + // ~/rpmbuild/BUILDROOT/foo-1.0.0-1.fc35.x86_64/usr/src/debug/foo-1.0.0-1.fc35.x86_64 + // directory. Given that these source files are actually located in + // the bpkg configuration directory rather than in the + // ~/rpmbuild/BUILD/foo-1.0.0 directory the cpio program fails to + // stat them and complains. To work around that we need to change + // the replacement directory path in the -ffile-prefix-map option + // value with some other absolute (not necessarily existing) path + // which is not a subdirectory of the directory prefixes recognized + // by the debugedit program. This way debugedit won't recognize any + // of the package source files, will create an empty + // debugsources.list file, and thus the cpio program won't try to + // copy anything. Not to confuse the user (who can potentially see + // such paths in gdb while examining a core file produced by the + // package binary), we will keep this replacement directory path + // close to the desired one, but will also make it clear that the + // path is bogus: // - // cpio: libfoo-1.2.3/foo.cxx: Cannot stat: No such file or directory + // /usr/src/debug/bogus/foo-1.0.0-1.fc35.x86_64 // - // This bloats the rpmbuild output but doesn't seem to break - // anything. + // Note that this path mapping won't work for external packages with + // source out of configuration (e.g., managed by bdep). + // + // @@ Supposedly this code won't be necessary when we add support + // for -debugsource sub-packages somehow. Feels like one way + // would be to make ~/rpmbuild/BUILD/foo-1.0.0 a symlink to the + // bpkg configuration (or to the primary package inside, if not + // --recursive). // if (ops_->fedora_buildflags () != "ignore") { + const char* debugsource_dir ( + "%{_usrsrc}/debug/bogus/%{name}-%{evr}.%{_arch}"); + if (lang_c || lang_cc) os << "%global build_cflags %{?build_cflags} -ffile-prefix-map=" - << cfg_dir.string () << "=." << '\n'; + << cfg_dir.string () << '=' << debugsource_dir << '\n'; if (lang_cxx || lang_cc) os << "%global build_cxxflags %{?build_cxxflags} -ffile-prefix-map=" - << cfg_dir.string () << "=." << '\n'; + << cfg_dir.string () << '=' << debugsource_dir << '\n'; } } else |