revset: use resolvehexnodeidprefix() in id() predicate (BC) We now have a public method for this purpose, so we don't need to access the private revlog._partialmatch(). Also, I'll probably make some changes to resolvehexnodeidprefix() later, and I want those to be reflected by the id() predicate. Note that this breaks a test case, because we now resolve the prefix in the unfiltered repo and get an ambiguous lookup, which results in no revision being added to the revset. The test case was already documented to be broken even though it wasn't. It's important to note that {shortest(node)} already uses the unfiltered repo, so we're not going to break people who get the prefix from there. I think we may not want to ever use shortest() in the filtered repo. It seems unlikely to be enough of a win to matter much. For example, in my hg repo, it would save me only 0.2 hex digits. In another repo that only I modify, it saves a little more, but it's still only 0.29 hex digits. It seems unlikely that people will prune enough commits that only 1/16 of the commits are visible (which is what it would take a to save a single hex digit). Instead, I'm working on another approach: allow ambiguous matches to be disambiguated within a user-specified revset. Whether or not that pans out, I hope we're okay with this little change in behavior for now and we can decide what to do about it later. Differential Revision: https://phab.mercurial-scm.org/D3311

revset: make id() an empty set for ambiguous nodeid (BC) As Yuya pointed out in the review of D3311, id() (and rev()) does not raise an error when the input is an unknown identifier, so it doesn't make sense for it to do that when the input is ambiguous with a filtered node. However, it turned out that it already does raise an error when the input is ambiguous among the visible nodes. So let's start by fixing that. Differential Revision: https://phab.mercurial-scm.org/D3462

shortest: don't keep checking for longer prefix if node doesn't exist (API) If revlog.shortest() is called with an invalid nodeid, we keep checking if longer and longer prefixes are valid. We call revlog._partialmatch() for each prefix. That function will give us None if the node doesn't exist (and a RevlogError if it's ambiguous), so there's no need to keep checking. This patch instead makes revlog.shortest() raise a LookupError is the node does not exist, and updates the caller to handle it. Before this patch, revlog.shortest() would return the full hexnode for nonexistent nodeids. By the same reasoning as in 7b2955624777 (scmutil: make shortesthexnodeidprefix() take a full binary nodeid, 2018-04-14), it's not revlog.shortest() that should decide how to present nonexistent nodeids, so that's now moved to the template function. This should speed up cases where {shortest()} is applied to an invalid nodeid, but I couldn't think of a reasonable case where that would happen. Differential Revision: https://phab.mercurial-scm.org/D3461

shortest: extract function for checking if a prefix is a revnum Much of isvalid() was about testing if a prefix is a valid revnum. I want to reuse that soon, so let's move it out. There is no significant slowdown from the function call overhead. Differential Revision: https://phab.mercurial-scm.org/D3460

shortest: rename "test" variable to "prefix" Sorry if this is considered churn, but "prefix" just seems much clearer to me. Differential Revision: https://phab.mercurial-scm.org/D3459

shortest: move some safe code out of exception block The RevlogError and WdirUnsupported could be raised by _partialmatch(), but not by the rest of isvalid(), so let's move the rest out to make it clearer. Differential Revision: https://phab.mercurial-scm.org/D3458

revlog: don't say "not found" on internal error If index_node() returned NULL, then index_find_node() and and nt_partialmatch() used to return -2 to signal that the node was not found. However, we were passing in a revnum to index_node() that we knew should exist, so the only reason it could return NULL was due to some internal error or perhaps out of memory. Let's not use "not found" for these cases. I suppose we never noticed this because these error never happen in practice. I think there are more places where we should error out instead of reporting that the node was not found, but the cases mentioned above were all I cared about right now (because using the same error code for all failures simplified some future patches). Differential Revision: https://phab.mercurial-scm.org/D3457