On Sun, Aug 26, 2018 at 10:27:10PM -0700, L A Walsh wrote: > when smartcols.sym has a whole section with 2.25 syms in it. > > What is the point of embedding mixing symbols with versions if they don't > work? ? > I.e. the 2.25 version of the symbols is in the lib, but findmnt linked > against 2.25 doesn't work cuz it can't access that specific label. (sources > from > 2.32.1). Not sure if I understand. findmnt with symbols from 2.25 will work with library from 2.32 as the library provides all previous versions too. For example: LD_LIBRARY_PATH=/home/projects/util-linux/util-linux/.libs/ /usr/bin/findmnt allows me to run old findmnt 2.30.2 (fedora 27) against library from current git tree (v2.33). > I don't understand why the symbols are versioned in a library that doesn't > appear to have anything other than a normal symbol table. * list of symbols defines the library API. It's better to have explicit list in the .sym file than depend on static/extern keywords in the code. * symbols versioning is fine grained solution. The library does not need to change library soname version; symbols versioning provides the same functionality on symbols level. * library is ready for future incompatible changes * versioned symbols allow to create explicit dependence between binary and library. This is used for example by distribution packaging systems (e.g. rpm) where package with binary requires specific symbols and package with library provides the symbols. It's better than dependence on library soname. * binary with new symbols will not executed against library with old symbols only. > So why are these symbols versioned if they don't work? They seem to prevent > earlier versions of programs from working with later libs even though they > are compatible. It's not true if the library provides old versions too. See excellent Ulrich's shared library howto for more details (chapter 3.3): https://software.intel.com/sites/default/files/m/a/1/e/dsohowto.pdf Karel -- Karel Zak <kzak@xxxxxxxxxx> http://karelzak.blogspot.com
