29.02.2016, 17:18, "Marc Lehmann": > On Mon, Feb 29, 2016 at 01:13:21PM +0300, Ponomarenko Andrey wrote: >> Hello Marc, >> >> The order of versions has been fixed: >> http://abi-laboratory.pro/tracker/timeline/libev/ > > 100% compatibility except on soname changes, as it should be (I am a bit > surprised, but not unhappy). > > I wonder how the percenmtages are calculated, though, isn't a changed soname > a 0% ABI backward compatibility, or are soname changes excepted? > > Also, I note that this ABI report seems to be solely about exported > symbols, other ABI changes are not checked (for example, when a symbol > in a header file changes it's value and/or meaning - at least the former > could be tracked automatically, although does not always result in an ABI > change). > > Or is this considered to be in the header diff? I see that the header diff > also shows differences in comments only (such as a year change in the > copyright notice), which doesn't look so helpful. > > Lastly, what about internal, header-only symbols? They might be exported, but > are not part of the official ABI. > > In any case, these questions are mostly meant for you - if all is withing > design parameters, then you don't have to bother with them and ignore me, I > don't have any outstanding issues with this tool.
Hello, Thank you for the review. The percentages are calculated by the abi-compliance-checker tool independently from analysis of the soname change. Changing the soname doens't always cause 0% compatibility. If you preserve a binary with the old soname in the system then compatibility is 100%. All symbols (not only exported by the object) can be analyzed if specifying additional -all-symbols option of the abi-dumper tool when creating ABI dumps for the library objects and specifying additional -source option of the abicc tool to check for source-level compatibility. Internal exported header-only symbols should be manually excluded from the report by specifying the list of official ABI symbols or a list of symbols to skip. Thank you. _______________________________________________ libev mailing list [email protected] http://lists.schmorp.de/mailman/listinfo/libev
