On Tue, May 06, 2008 at 10:17:32PM -0700, David R. Morrison wrote: > I've tried to write an expanded error message for the situation where > the validator encounters a shared library in a .deb which is not > listed in the Shlibs field. (The code is in HEAD, but not yet > released.) Here's a sample error message:
Yeah, we need to make people understand it. OTOH, the more detailed we make this message, the more special-cases slip through... > Validating .deb file fontforge_20080330-1_darwin-powerpc.deb... > Error: package contains a shared library > /sw/lib/fontforge/libfontforge.1.0.0.dylib > but the install_name and compatibility_version > (%p/lib/fontforge/libfontforge.1.dylib 2.0.0) > are not listed in the Shlibs field. If this library is > public, in the > sense that it may be linked to by other packages, then these > must be > listed in the Shlibs field, along with versioning information > about the > package providing the library. However, if this is a private > library > which will be used only by this package, add > '!%p/lib/fontforge/libfontforge.1.dylib' > to the Shlibs field. "the install_name and compatibility_version [...] are not listed in the Shlibs field." But then later is an instruction to list *only* the install_name (not c_v). Also, don't want to parenthesize the i_n and c_v, since it suggests that that's the Shlibs: syntax (likewise, no single-quotes on the private-shlibs syntax). > (Note to Daniel: your package doesn't actually generate this error: I > hacked it so that it would!) > > I hope that this both correctly and clearly conveys the intent -- I > know that earlier versions have had shortcomings. > > Also, if the install_name doesn't begin with /sw (or your local > prefix), a different error message is generated which points out that > the install_name is malformed. This is to address the apparent issue > reported here: http://www.nabble.com/qmake,-no-.la,-and-a-validator- > error-td16228400.html and here: http://www.opensubscriber.com/message/ > fink-devel@lists.sourceforge.net/8879391.html The code in HEAD does a regex match for %p (which is good--better to use that literal in the Shlibs info instead of hardcoding /sw or whatever in the error msgs) but there are legit packages that install outside of %p and still need standard Shlibs handling. Not quite sure what the else{} block is supposed to catch. I *think* it traps install_name that isn't an absolute path? Need to anchor the regex pattern to the beginning of the string, and it's still leaky (especially for a libtool or hand-coded-alternative screwup) if it points to the build dir (since that's a subdir of %p). Would it be better to do the main message if it's *any* absolute path (i.e., not just in $basepath) but still s/^basepath/%p/ for the displayed message and the else{} if a non-absolute one? dan -- Daniel Macks [EMAIL PROTECTED] http://www.netspace.org/~dmacks ------------------------------------------------------------------------- This SF.net email is sponsored by the 2008 JavaOne(SM) Conference Don't miss this year's exciting event. There's still time to save $100. Use priority code J8TL2D2. http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone _______________________________________________ Fink-devel mailing list Fink-devel@lists.sourceforge.net http://news.gmane.org/gmane.os.apple.fink.devel
