On 26.12.2018 23:44, Daniel Shahaf wrote: > Branko Čibej wrote on Wed, 26 Dec 2018 23:37 +0100: >> On 26.12.2018 23:21, Daniel Shahaf wrote: >>> Branko Čibej wrote on Wed, 26 Dec 2018 22:41 +0100: >>>> On 26.12.2018 19:50, Daniel Shahaf wrote: >>>>> Haven't reviewed the rest of the patch, nor the mapping of >>>>> svn_error_t::apr_err values to this hierarchy. >>>> There is just one exception type that encapsulates all of svn_error_t, >>>> including the apr_err bit; that's 'svn::error'. I have no intention of >>>> going down the rabbit hole of having one exception type for each >>>> possible apr_err value! >>>> >>> Yeah, that'd be way too much; but I was thinking of two things: >>> >>> 1. apr_err can be an errno error code, not just an APR_OS_START_USERERR >>> code. I don't have the C++ exceptions hierarchy in mind, but I >>> suspect that when APR_STATUS_IS_EEXIST(err->apr_err), an svn::error >>> instance is not what people (and 'catch' blocks) will expect. >> >> Ah, good point. I hadn't actually thought about this side of things, as >> clients "shouldn't" have to worry about them iff our error messages make >> any sense. But, yes, adding such predicates would be a big help. >> >> They don't actually have to be part of svn::error, I'd make them >> namespace-scope functions, e.g.: >> >> bool svn::error_is_eexist(const svn::error& e) noexcept; >> >> >> the point being that the svn::error object serves as both an exception >> type and a detailed error description (and that's the reason for >> deriving svn::cancelled from svn::error). > I don't disagree with adding such predicates, but they aren't my point. > What I was trying to say is, doesn't the standard C++ exception > hierarchy have some std::* exception class for, say, IO errors? In > which case, C++ consumers might be surprised that an svn::error object > with apr_err==EEXIST isn't caught by their 'catch' clauses for std::* IO > errors?
Not so much. Or at least not as detailed as you'd think: https://en.cppreference.com/w/cpp/error/exception They'll expect std::ios_base::failure from one of the stream functions, but not from Subversion. and std::filesystem::filesystem_error is specific to the std::filesystem library, which we can't use in our API because it's C++17 -- and even if we could use it conditionally, it would be bad form to reuse those errors for conditions that did not arise from use of the std::filesystem API. I'm already changing svn::allocation_failed to be derived from std::bad_alloc instead of std::runtime_error, but for the rest, it's best to leave well enough alone. Users _will_ have to implement specific error handling for the SVN++ parts of their code, or if they don't, just rely on catch-all behaviour that may or may not lose error context. -- Brane
