From 5dd6c4f062ec392b965a5fcfc18fd3f75e2a79cb Mon Sep 17 00:00:00 2001 From: Robert Hensing Date: Mon, 19 Aug 2024 13:21:44 +0200 Subject: [PATCH 1/8] libgit2, GitRepo: Write thin packfiles libgit2 didn't write thin ones, hence the patch. This should improve performance on systems with weak I/O in ~/.cache, especially in terms of operations per second, or where system calls are slower. (macOS, VMs?) --- packaging/dependencies.nix | 2 + .../libgit2-mempack-thin-packfile.patch | 282 ++++++++++++++++++ src/libfetchers/git-utils.cc | 57 ++++ src/libfetchers/git-utils.hh | 2 + 4 files changed, 343 insertions(+) create mode 100644 packaging/patches/libgit2-mempack-thin-packfile.patch diff --git a/packaging/dependencies.nix b/packaging/dependencies.nix index 21c48e5cc..74b5cbc05 100644 --- a/packaging/dependencies.nix +++ b/packaging/dependencies.nix @@ -115,6 +115,8 @@ scope: { version = inputs.libgit2.lastModifiedDate; cmakeFlags = attrs.cmakeFlags or [] ++ [ "-DUSE_SSH=exec" ]; + patches = attrs.patches or [] + ++ [ ./patches/libgit2-mempack-thin-packfile.patch ]; }); busybox-sandbox-shell = pkgs.busybox-sandbox-shell or (pkgs.busybox.override { diff --git a/packaging/patches/libgit2-mempack-thin-packfile.patch b/packaging/patches/libgit2-mempack-thin-packfile.patch new file mode 100644 index 000000000..fb74b1683 --- /dev/null +++ b/packaging/patches/libgit2-mempack-thin-packfile.patch @@ -0,0 +1,282 @@ +commit 9bacade4a3ef4b6b26e2c02f549eef0e9eb9eaa2 +Author: Robert Hensing +Date: Sun Aug 18 20:20:36 2024 +0200 + + Add unoptimized git_mempack_write_thin_pack + +diff --git a/include/git2/sys/mempack.h b/include/git2/sys/mempack.h +index 17da590a3..3688bdd50 100644 +--- a/include/git2/sys/mempack.h ++++ b/include/git2/sys/mempack.h +@@ -44,6 +44,29 @@ GIT_BEGIN_DECL + */ + GIT_EXTERN(int) git_mempack_new(git_odb_backend **out); + ++/** ++ * Write a thin packfile with the objects in the memory store. ++ * ++ * A thin packfile is a packfile that does not contain its transitive closure of ++ * references. This is useful for efficiently distributing additions to a ++ * repository over the network, but also finds use in the efficient bulk ++ * addition of objects to a repository, locally. ++ * ++ * This operation performs the (shallow) insert operations into the ++ * `git_packbuilder`, but does not write the packfile to disk; ++ * see `git_packbuilder_write_buf`. ++ * ++ * It also does not reset the memory store; see `git_mempack_reset`. ++ * ++ * @note This function may or may not write trees and blobs that are not ++ * referenced by commits. Currently everything is written, but this ++ * behavior may change in the future as the packer is optimized. ++ * ++ * @param backend The mempack backend ++ * @param pb The packbuilder to use to write the packfile ++ */ ++GIT_EXTERN(int) git_mempack_write_thin_pack(git_odb_backend *backend, git_packbuilder *pb); ++ + /** + * Dump all the queued in-memory writes to a packfile. + * +diff --git a/src/libgit2/odb_mempack.c b/src/libgit2/odb_mempack.c +index 6f27f45f8..0b61e2b66 100644 +--- a/src/libgit2/odb_mempack.c ++++ b/src/libgit2/odb_mempack.c +@@ -132,6 +132,35 @@ cleanup: + return err; + } + ++int git_mempack_write_thin_pack(git_odb_backend *backend, git_packbuilder *pb) ++{ ++ struct memory_packer_db *db = (struct memory_packer_db *)backend; ++ const git_oid *oid; ++ size_t iter = 0; ++ int err = -1; ++ ++ /* TODO: Implement the recency heuristics. ++ For this it probably makes sense to only write what's referenced ++ through commits, an option I've carved out for you in the docs. ++ wrt heuristics: ask your favorite LLM to translate https://git-scm.com/docs/pack-heuristics/en ++ to actual normal reference documentation. */ ++ while (true) { ++ err = git_oidmap_iterate(NULL, db->objects, &iter, &oid); ++ if (err == GIT_ITEROVER) { ++ err = 0; ++ break; ++ } ++ if (err != 0) ++ return err; ++ ++ err = git_packbuilder_insert(pb, oid, NULL); ++ if (err != 0) ++ return err; ++ } ++ ++ return 0; ++} ++ + int git_mempack_dump( + git_buf *pack, + git_repository *repo, +diff --git a/tests/libgit2/mempack/thinpack.c b/tests/libgit2/mempack/thinpack.c +new file mode 100644 +index 000000000..604a4dda2 +--- /dev/null ++++ b/tests/libgit2/mempack/thinpack.c +@@ -0,0 +1,196 @@ ++#include "clar_libgit2.h" ++#include "git2/indexer.h" ++#include "git2/odb_backend.h" ++#include "git2/tree.h" ++#include "git2/types.h" ++#include "git2/sys/mempack.h" ++#include "git2/sys/odb_backend.h" ++#include "util.h" ++ ++static git_repository *_repo; ++static git_odb_backend * _mempack_backend; ++ ++void test_mempack_thinpack__initialize(void) ++{ ++ git_odb *odb; ++ ++ _repo = cl_git_sandbox_init_new("mempack_thinpack_repo"); ++ ++ cl_git_pass(git_mempack_new(&_mempack_backend)); ++ cl_git_pass(git_repository_odb(&odb, _repo)); ++ cl_git_pass(git_odb_add_backend(odb, _mempack_backend, 999)); ++ git_odb_free(odb); ++} ++ ++void _mempack_thinpack__cleanup(void) ++{ ++ cl_git_sandbox_cleanup(); ++} ++ ++/* ++ Generating a packfile for an unchanged repo works and produces an empty packfile. ++ Even if we allow this scenario to be detected, it shouldn't misbehave if the ++ application is unaware of it. ++*/ ++void test_mempack_thinpack__empty(void) ++{ ++ git_packbuilder *pb; ++ int version; ++ int n; ++ git_buf buf = GIT_BUF_INIT; ++ ++ git_packbuilder_new(&pb, _repo); ++ ++ cl_git_pass(git_mempack_write_thin_pack(_mempack_backend, pb)); ++ cl_git_pass(git_packbuilder_write_buf(&buf, pb)); ++ cl_assert_in_range(12, buf.size, 1024 /* empty packfile is >0 bytes, but certainly not that big */); ++ cl_assert(buf.ptr[0] == 'P'); ++ cl_assert(buf.ptr[1] == 'A'); ++ cl_assert(buf.ptr[2] == 'C'); ++ cl_assert(buf.ptr[3] == 'K'); ++ version = (buf.ptr[4] << 24) | (buf.ptr[5] << 16) | (buf.ptr[6] << 8) | buf.ptr[7]; ++ /* Subject to change. https://git-scm.com/docs/pack-format: Git currently accepts version number 2 or 3 but generates version 2 only.*/ ++ cl_assert_equal_i(2, version); ++ n = (buf.ptr[8] << 24) | (buf.ptr[9] << 16) | (buf.ptr[10] << 8) | buf.ptr[11]; ++ cl_assert_equal_i(0, n); ++ git_buf_dispose(&buf); ++ ++ git_packbuilder_free(pb); ++} ++ ++#define LIT_LEN(x) x, sizeof(x) - 1 ++ ++/* ++ Check that git_mempack_write_thin_pack produces a thin packfile. ++*/ ++void test_mempack_thinpack__thin(void) ++{ ++ /* Outline: ++ - Create tree 1 ++ - Flush to packfile A ++ - Create tree 2 ++ - Flush to packfile B ++ ++ Tree 2 has a new blob and a reference to a blob from tree 1. ++ ++ Expectation: ++ - Packfile B is thin and does not contain the objects from packfile A ++ */ ++ ++ ++ git_oid oid_blob_1; ++ git_oid oid_blob_2; ++ git_oid oid_blob_3; ++ git_oid oid_tree_1; ++ git_oid oid_tree_2; ++ git_treebuilder *tb; ++ ++ git_packbuilder *pb; ++ git_buf buf = GIT_BUF_INIT; ++ git_indexer *indexer; ++ git_indexer_progress stats; ++ char pack_dir_path[1024]; ++ ++ char sbuf[1024]; ++ const char * repo_path; ++ const char * pack_name_1; ++ const char * pack_name_2; ++ git_str pack_path_1 = GIT_STR_INIT; ++ git_str pack_path_2 = GIT_STR_INIT; ++ git_odb_backend * pack_odb_backend_1; ++ git_odb_backend * pack_odb_backend_2; ++ ++ ++ cl_assert_in_range(0, snprintf(pack_dir_path, sizeof(pack_dir_path), "%s/objects/pack", git_repository_path(_repo)), sizeof(pack_dir_path)); ++ ++ /* Create tree 1 */ ++ ++ cl_git_pass(git_blob_create_from_buffer(&oid_blob_1, _repo, LIT_LEN("thinpack blob 1"))); ++ cl_git_pass(git_blob_create_from_buffer(&oid_blob_2, _repo, LIT_LEN("thinpack blob 2"))); ++ ++ ++ cl_git_pass(git_treebuilder_new(&tb, _repo, NULL)); ++ cl_git_pass(git_treebuilder_insert(NULL, tb, "blob1", &oid_blob_1, GIT_FILEMODE_BLOB)); ++ cl_git_pass(git_treebuilder_insert(NULL, tb, "blob2", &oid_blob_2, GIT_FILEMODE_BLOB)); ++ cl_git_pass(git_treebuilder_write(&oid_tree_1, tb)); ++ ++ /* Flush */ ++ ++ cl_git_pass(git_packbuilder_new(&pb, _repo)); ++ cl_git_pass(git_mempack_write_thin_pack(_mempack_backend, pb)); ++ cl_git_pass(git_packbuilder_write_buf(&buf, pb)); ++ cl_git_pass(git_indexer_new(&indexer, pack_dir_path, 0, NULL, NULL)); ++ cl_git_pass(git_indexer_append(indexer, buf.ptr, buf.size, &stats)); ++ cl_git_pass(git_indexer_commit(indexer, &stats)); ++ pack_name_1 = strdup(git_indexer_name(indexer)); ++ cl_assert(pack_name_1); ++ git_buf_dispose(&buf); ++ git_mempack_reset(_mempack_backend); ++ git_indexer_free(indexer); ++ git_packbuilder_free(pb); ++ ++ /* Create tree 2 */ ++ ++ cl_git_pass(git_treebuilder_clear(tb)); ++ /* blob 1 won't be used, but we add it anyway to test that just "declaring" an object doesn't ++ necessarily cause its inclusion in the next thin packfile. It must only be included if new. */ ++ cl_git_pass(git_blob_create_from_buffer(&oid_blob_1, _repo, LIT_LEN("thinpack blob 1"))); ++ cl_git_pass(git_blob_create_from_buffer(&oid_blob_3, _repo, LIT_LEN("thinpack blob 3"))); ++ cl_git_pass(git_treebuilder_insert(NULL, tb, "blob1", &oid_blob_1, GIT_FILEMODE_BLOB)); ++ cl_git_pass(git_treebuilder_insert(NULL, tb, "blob3", &oid_blob_3, GIT_FILEMODE_BLOB)); ++ cl_git_pass(git_treebuilder_write(&oid_tree_2, tb)); ++ ++ /* Flush */ ++ ++ cl_git_pass(git_packbuilder_new(&pb, _repo)); ++ cl_git_pass(git_mempack_write_thin_pack(_mempack_backend, pb)); ++ cl_git_pass(git_packbuilder_write_buf(&buf, pb)); ++ cl_git_pass(git_indexer_new(&indexer, pack_dir_path, 0, NULL, NULL)); ++ cl_git_pass(git_indexer_append(indexer, buf.ptr, buf.size, &stats)); ++ cl_git_pass(git_indexer_commit(indexer, &stats)); ++ pack_name_2 = strdup(git_indexer_name(indexer)); ++ cl_assert(pack_name_2); ++ git_buf_dispose(&buf); ++ git_mempack_reset(_mempack_backend); ++ git_indexer_free(indexer); ++ git_packbuilder_free(pb); ++ git_treebuilder_free(tb); ++ ++ /* Assertions */ ++ ++ assert(pack_name_1); ++ assert(pack_name_2); ++ ++ repo_path = git_repository_path(_repo); ++ ++ snprintf(sbuf, sizeof(sbuf), "objects/pack/pack-%s.pack", pack_name_1); ++ git_str_joinpath(&pack_path_1, repo_path, sbuf); ++ snprintf(sbuf, sizeof(sbuf), "objects/pack/pack-%s.pack", pack_name_2); ++ git_str_joinpath(&pack_path_2, repo_path, sbuf); ++ ++ /* If they're the same, something definitely went wrong. */ ++ cl_assert(strcmp(pack_name_1, pack_name_2) != 0); ++ ++ cl_git_pass(git_odb_backend_one_pack(&pack_odb_backend_1, pack_path_1.ptr)); ++ cl_assert(pack_odb_backend_1->exists(pack_odb_backend_1, &oid_blob_1)); ++ cl_assert(pack_odb_backend_1->exists(pack_odb_backend_1, &oid_blob_2)); ++ cl_assert(!pack_odb_backend_1->exists(pack_odb_backend_1, &oid_blob_3)); ++ cl_assert(pack_odb_backend_1->exists(pack_odb_backend_1, &oid_tree_1)); ++ cl_assert(!pack_odb_backend_1->exists(pack_odb_backend_1, &oid_tree_2)); ++ ++ cl_git_pass(git_odb_backend_one_pack(&pack_odb_backend_2, pack_path_2.ptr)); ++ /* blob 1 is already in the packfile 1, so packfile 2 must not include it, in order to be _thin_. */ ++ cl_assert(!pack_odb_backend_2->exists(pack_odb_backend_2, &oid_blob_1)); ++ cl_assert(!pack_odb_backend_2->exists(pack_odb_backend_2, &oid_blob_2)); ++ cl_assert(pack_odb_backend_2->exists(pack_odb_backend_2, &oid_blob_3)); ++ cl_assert(!pack_odb_backend_2->exists(pack_odb_backend_2, &oid_tree_1)); ++ cl_assert(pack_odb_backend_2->exists(pack_odb_backend_2, &oid_tree_2)); ++ ++ pack_odb_backend_1->free(pack_odb_backend_1); ++ pack_odb_backend_2->free(pack_odb_backend_2); ++ free((void *)pack_name_1); ++ free((void *)pack_name_2); ++ git_str_dispose(&pack_path_1); ++ git_str_dispose(&pack_path_2); ++ ++} diff --git a/src/libfetchers/git-utils.cc b/src/libfetchers/git-utils.cc index 114aa4ec0..5306d8780 100644 --- a/src/libfetchers/git-utils.cc +++ b/src/libfetchers/git-utils.cc @@ -13,13 +13,17 @@ #include #include #include +#include #include +#include #include #include #include #include #include #include +#include +#include #include #include @@ -76,6 +80,9 @@ typedef std::unique_ptr> StatusLi typedef std::unique_ptr> Remote; typedef std::unique_ptr> GitConfig; typedef std::unique_ptr> ConfigIterator; +typedef std::unique_ptr> ObjectDb; +typedef std::unique_ptr> PackBuilder; +typedef std::unique_ptr> Indexer; // A helper to ensure that we don't leak objects returned by libgit2. template @@ -164,6 +171,11 @@ struct GitRepoImpl : GitRepo, std::enable_shared_from_this /** Location of the repository on disk. */ std::filesystem::path path; Repository repo; + /** + * In-memory object store for efficient batched writing to packfiles. + * Owned by `repo`. + */ + git_odb_backend * mempack_backend; GitRepoImpl(std::filesystem::path _path, bool create, bool bare) : path(std::move(_path)) @@ -177,6 +189,16 @@ struct GitRepoImpl : GitRepo, std::enable_shared_from_this if (git_repository_init(Setter(repo), path.string().c_str(), bare)) throw Error("creating Git repository '%s': %s", path, git_error_last()->message); } + git_odb * odb; + if (git_repository_odb(&odb, repo.get())) + throw Error("getting Git object database: %s", git_error_last()->message); + + // TODO: release mempack_backend? + if (git_mempack_new(&mempack_backend)) + throw Error("creating mempack backend: %s", git_error_last()->message); + + if (git_odb_add_backend(odb, mempack_backend, 999)) + throw Error("adding mempack backend to Git object database: %s", git_error_last()->message); } operator git_repository * () @@ -184,6 +206,39 @@ struct GitRepoImpl : GitRepo, std::enable_shared_from_this return repo.get(); } + void flush() override { + git_buf buf = GIT_BUF_INIT; + try { + PackBuilder packBuilder; + git_packbuilder_new(Setter(packBuilder), *this); + git_mempack_write_thin_pack(mempack_backend, packBuilder.get()); + git_packbuilder_write_buf(&buf, packBuilder.get()); + + std::string repo_path = std::string(git_repository_path(repo.get())); + while (!repo_path.empty() && repo_path.back() == '/') + repo_path.pop_back(); + std::string pack_dir_path = repo_path + "/objects/pack"; + + // TODO: could the indexing be done in a separate thread? + Indexer indexer; + git_indexer_progress stats; + if (git_indexer_new(Setter(indexer), pack_dir_path.c_str(), 0, nullptr, nullptr)) + throw Error("creating git packfile indexer: %s", git_error_last()->message); + if (git_indexer_append(indexer.get(), buf.ptr, buf.size, &stats)) + throw Error("appending to git packfile index: %s", git_error_last()->message); + if (git_indexer_commit(indexer.get(), &stats)) + throw Error("committing git packfile index: %s", git_error_last()->message); + + if (git_mempack_reset(mempack_backend)) + throw Error("resetting git mempack backend: %s", git_error_last()->message); + + git_buf_dispose(&buf); + } catch (...) { + git_buf_dispose(&buf); + throw; + } + } + uint64_t getRevCount(const Hash & rev) override { std::unordered_set done; @@ -1008,6 +1063,8 @@ struct GitFileSystemObjectSinkImpl : GitFileSystemObjectSink auto [oid, _name] = popBuilder(); + repo->flush(); + return toHash(oid); } }; diff --git a/src/libfetchers/git-utils.hh b/src/libfetchers/git-utils.hh index 915252868..65a598ce5 100644 --- a/src/libfetchers/git-utils.hh +++ b/src/libfetchers/git-utils.hh @@ -80,6 +80,8 @@ struct GitRepo virtual ref getFileSystemObjectSink() = 0; + virtual void flush() = 0; + virtual void fetch( const std::string & url, const std::string & refspec, From d0f8a9236392ab41414dddbe7400da98fad9f62d Mon Sep 17 00:00:00 2001 From: Robert Hensing Date: Mon, 26 Aug 2024 11:38:40 +0200 Subject: [PATCH 2/8] Make tarball cache more interruptible --- src/libfetchers/git-utils.cc | 7 +++++++ src/libutil/unix/signals-impl.hh | 6 ++++++ 2 files changed, 13 insertions(+) diff --git a/src/libfetchers/git-utils.cc b/src/libfetchers/git-utils.cc index 5306d8780..fb5341599 100644 --- a/src/libfetchers/git-utils.cc +++ b/src/libfetchers/git-utils.cc @@ -211,8 +211,12 @@ struct GitRepoImpl : GitRepo, std::enable_shared_from_this try { PackBuilder packBuilder; git_packbuilder_new(Setter(packBuilder), *this); + checkInterrupt(); git_mempack_write_thin_pack(mempack_backend, packBuilder.get()); + checkInterrupt(); + // TODO make git_packbuilder_write_buf() interruptible git_packbuilder_write_buf(&buf, packBuilder.get()); + checkInterrupt(); std::string repo_path = std::string(git_repository_path(repo.get())); while (!repo_path.empty() && repo_path.back() == '/') @@ -224,8 +228,10 @@ struct GitRepoImpl : GitRepo, std::enable_shared_from_this git_indexer_progress stats; if (git_indexer_new(Setter(indexer), pack_dir_path.c_str(), 0, nullptr, nullptr)) throw Error("creating git packfile indexer: %s", git_error_last()->message); + // TODO: feed buf in (fairly large) chunk to make this interruptible if (git_indexer_append(indexer.get(), buf.ptr, buf.size, &stats)) throw Error("appending to git packfile index: %s", git_error_last()->message); + checkInterrupt(); if (git_indexer_commit(indexer.get(), &stats)) throw Error("committing git packfile index: %s", git_error_last()->message); @@ -237,6 +243,7 @@ struct GitRepoImpl : GitRepo, std::enable_shared_from_this git_buf_dispose(&buf); throw; } + checkInterrupt(); } uint64_t getRevCount(const Hash & rev) override diff --git a/src/libutil/unix/signals-impl.hh b/src/libutil/unix/signals-impl.hh index 7ac8c914d..2193922be 100644 --- a/src/libutil/unix/signals-impl.hh +++ b/src/libutil/unix/signals-impl.hh @@ -84,6 +84,12 @@ static inline bool getInterrupted() return unix::_isInterrupted; } +/** + * Throw `Interrupted` exception if the process has been interrupted. + * + * Call this in long-running loops and between slow operations to terminate + * them as needed. + */ void inline checkInterrupt() { using namespace unix; From 97ff2ed4555ce0761368046270fc4466550bfb0c Mon Sep 17 00:00:00 2001 From: Robert Hensing Date: Mon, 26 Aug 2024 11:39:14 +0200 Subject: [PATCH 3/8] Sync tarball cache within tarball cache Activity --- src/libfetchers/github.cc | 3 ++- src/libfetchers/tarball.cc | 3 ++- 2 files changed, 4 insertions(+), 2 deletions(-) diff --git a/src/libfetchers/github.cc b/src/libfetchers/github.cc index 2e914164a..ecfd035fc 100644 --- a/src/libfetchers/github.cc +++ b/src/libfetchers/github.cc @@ -261,11 +261,12 @@ struct GitArchiveInputScheme : InputScheme auto tarballCache = getTarballCache(); auto parseSink = tarballCache->getFileSystemObjectSink(); auto lastModified = unpackTarfileToSink(archive, *parseSink); + auto tree = parseSink->sync(); act.reset(); TarballInfo tarballInfo { - .treeHash = tarballCache->dereferenceSingletonDirectory(parseSink->sync()), + .treeHash = tarballCache->dereferenceSingletonDirectory(tree), .lastModified = lastModified }; diff --git a/src/libfetchers/tarball.cc b/src/libfetchers/tarball.cc index dd4f3b780..a082b0078 100644 --- a/src/libfetchers/tarball.cc +++ b/src/libfetchers/tarball.cc @@ -170,6 +170,7 @@ static DownloadTarballResult downloadTarball_( auto tarballCache = getTarballCache(); auto parseSink = tarballCache->getFileSystemObjectSink(); auto lastModified = unpackTarfileToSink(archive, *parseSink); + auto tree = parseSink->sync(); act.reset(); @@ -184,7 +185,7 @@ static DownloadTarballResult downloadTarball_( } else { infoAttrs.insert_or_assign("etag", res->etag); infoAttrs.insert_or_assign("treeHash", - tarballCache->dereferenceSingletonDirectory(parseSink->sync()).gitRev()); + tarballCache->dereferenceSingletonDirectory(tree).gitRev()); infoAttrs.insert_or_assign("lastModified", uint64_t(lastModified)); if (res->immutableUrl) infoAttrs.insert_or_assign("immutableUrl", *res->immutableUrl); From fb8d3ed1506d2c0e3065abaad9587e16dccfd1b9 Mon Sep 17 00:00:00 2001 From: Robert Hensing Date: Mon, 26 Aug 2024 15:04:47 +0200 Subject: [PATCH 4/8] fixup: sync -> flush The latter is not used for memory synchronization things. --- src/libfetchers/git-utils.cc | 7 ++++++- src/libfetchers/git-utils.hh | 6 +++++- src/libfetchers/github.cc | 2 +- src/libfetchers/tarball.cc | 2 +- tests/unit/libfetchers/git-utils.cc | 2 +- 5 files changed, 14 insertions(+), 5 deletions(-) diff --git a/src/libfetchers/git-utils.cc b/src/libfetchers/git-utils.cc index fb5341599..59b4d424a 100644 --- a/src/libfetchers/git-utils.cc +++ b/src/libfetchers/git-utils.cc @@ -170,6 +170,11 @@ struct GitRepoImpl : GitRepo, std::enable_shared_from_this { /** Location of the repository on disk. */ std::filesystem::path path; + /** + * libgit2 repository. Note that new objects are not written to disk, + * because we are using a mempack backend. For writing to disk, see + * `flush()`, which is also called by `GitFileSystemObjectSink::sync()`. + */ Repository repo; /** * In-memory object store for efficient batched writing to packfiles. @@ -1064,7 +1069,7 @@ struct GitFileSystemObjectSinkImpl : GitFileSystemObjectSink git_tree_entry_filemode(entry)); } - Hash sync() override + Hash flush() override { updateBuilders({}); diff --git a/src/libfetchers/git-utils.hh b/src/libfetchers/git-utils.hh index 65a598ce5..f45b5a504 100644 --- a/src/libfetchers/git-utils.hh +++ b/src/libfetchers/git-utils.hh @@ -7,12 +7,16 @@ namespace nix { namespace fetchers { struct PublicKey; } +/** + * A sink that writes into a Git repository. Note that nothing may be written + * until `flush()` is called. + */ struct GitFileSystemObjectSink : ExtendedFileSystemObjectSink { /** * Flush builder and return a final Git hash. */ - virtual Hash sync() = 0; + virtual Hash flush() = 0; }; struct GitRepo diff --git a/src/libfetchers/github.cc b/src/libfetchers/github.cc index ecfd035fc..308cff33a 100644 --- a/src/libfetchers/github.cc +++ b/src/libfetchers/github.cc @@ -261,7 +261,7 @@ struct GitArchiveInputScheme : InputScheme auto tarballCache = getTarballCache(); auto parseSink = tarballCache->getFileSystemObjectSink(); auto lastModified = unpackTarfileToSink(archive, *parseSink); - auto tree = parseSink->sync(); + auto tree = parseSink->flush(); act.reset(); diff --git a/src/libfetchers/tarball.cc b/src/libfetchers/tarball.cc index a082b0078..aa5d61bc5 100644 --- a/src/libfetchers/tarball.cc +++ b/src/libfetchers/tarball.cc @@ -170,7 +170,7 @@ static DownloadTarballResult downloadTarball_( auto tarballCache = getTarballCache(); auto parseSink = tarballCache->getFileSystemObjectSink(); auto lastModified = unpackTarfileToSink(archive, *parseSink); - auto tree = parseSink->sync(); + auto tree = parseSink->flush(); act.reset(); diff --git a/tests/unit/libfetchers/git-utils.cc b/tests/unit/libfetchers/git-utils.cc index de5110cc3..0bf3076dc 100644 --- a/tests/unit/libfetchers/git-utils.cc +++ b/tests/unit/libfetchers/git-utils.cc @@ -77,7 +77,7 @@ TEST_F(GitUtilsTest, sink_basic) // sink->createHardlink("foo-1.1/links/foo-2", CanonPath("foo-1.1/hello")); - auto result = repo->dereferenceSingletonDirectory(sink->sync()); + auto result = repo->dereferenceSingletonDirectory(sink->flush()); auto accessor = repo->getAccessor(result, false); auto entries = accessor->readDirectory(CanonPath::root); ASSERT_EQ(entries.size(), 5); From 57c48304bba7d1f92fb9acc8abc9e8db4450fc0c Mon Sep 17 00:00:00 2001 From: Robert Hensing Date: Mon, 26 Aug 2024 15:34:35 +0200 Subject: [PATCH 5/8] fixup: Release odb --- src/libfetchers/git-utils.cc | 9 +++++---- 1 file changed, 5 insertions(+), 4 deletions(-) diff --git a/src/libfetchers/git-utils.cc b/src/libfetchers/git-utils.cc index 59b4d424a..70391a287 100644 --- a/src/libfetchers/git-utils.cc +++ b/src/libfetchers/git-utils.cc @@ -194,15 +194,16 @@ struct GitRepoImpl : GitRepo, std::enable_shared_from_this if (git_repository_init(Setter(repo), path.string().c_str(), bare)) throw Error("creating Git repository '%s': %s", path, git_error_last()->message); } - git_odb * odb; - if (git_repository_odb(&odb, repo.get())) + + ObjectDb odb; + if (git_repository_odb(Setter(odb), repo.get())) throw Error("getting Git object database: %s", git_error_last()->message); - // TODO: release mempack_backend? + // mempack_backend will be owned by the repository, so we are not expected to free it ourselves. if (git_mempack_new(&mempack_backend)) throw Error("creating mempack backend: %s", git_error_last()->message); - if (git_odb_add_backend(odb, mempack_backend, 999)) + if (git_odb_add_backend(odb.get(), mempack_backend, 999)) throw Error("adding mempack backend to Git object database: %s", git_error_last()->message); } From c1fe3546ed67babeecfb376315063f0642ce4bbd Mon Sep 17 00:00:00 2001 From: Robert Hensing Date: Wed, 28 Aug 2024 01:48:25 +0200 Subject: [PATCH 6/8] libgit2: Add libgit2-packbuilder-callback-interruptible.patch --- packaging/dependencies.nix | 18 +- ...2-packbuilder-callback-interruptible.patch | 930 ++++++++++++++++++ 2 files changed, 947 insertions(+), 1 deletion(-) create mode 100644 packaging/patches/libgit2-packbuilder-callback-interruptible.patch diff --git a/packaging/dependencies.nix b/packaging/dependencies.nix index 74b5cbc05..0182f29c0 100644 --- a/packaging/dependencies.nix +++ b/packaging/dependencies.nix @@ -115,8 +115,24 @@ scope: { version = inputs.libgit2.lastModifiedDate; cmakeFlags = attrs.cmakeFlags or [] ++ [ "-DUSE_SSH=exec" ]; + nativeBuildInputs = attrs.nativeBuildInputs or [] + ++ [ + # Needed for `git apply`; see `prePatch` + pkgs.buildPackages.gitMinimal + ]; + # Only `git apply` can handle git binary patches + prePatch = '' + patch() { + git apply + } + ''; patches = attrs.patches or [] - ++ [ ./patches/libgit2-mempack-thin-packfile.patch ]; + ++ [ + ./patches/libgit2-mempack-thin-packfile.patch + + # binary patch; see `prePatch` + ./patches/libgit2-packbuilder-callback-interruptible.patch + ]; }); busybox-sandbox-shell = pkgs.busybox-sandbox-shell or (pkgs.busybox.override { diff --git a/packaging/patches/libgit2-packbuilder-callback-interruptible.patch b/packaging/patches/libgit2-packbuilder-callback-interruptible.patch new file mode 100644 index 000000000..c67822ff7 --- /dev/null +++ b/packaging/patches/libgit2-packbuilder-callback-interruptible.patch @@ -0,0 +1,930 @@ +commit e9823c5da4fa977c46bcb97167fbdd0d70adb5ff +Author: Robert Hensing +Date: Mon Aug 26 20:07:04 2024 +0200 + + Make packbuilder interruptible using progress callback + + Forward errors from packbuilder->progress_cb + + This allows the callback to terminate long-running operations when + the application is interrupted. + +diff --git a/include/git2/pack.h b/include/git2/pack.h +index 0f6bd2ab9..bee72a6c0 100644 +--- a/include/git2/pack.h ++++ b/include/git2/pack.h +@@ -247,6 +247,9 @@ typedef int GIT_CALLBACK(git_packbuilder_progress)( + * @param progress_cb Function to call with progress information during + * pack building. Be aware that this is called inline with pack building + * operations, so performance may be affected. ++ * When progress_cb returns an error, the pack building process will be ++ * aborted and the error will be returned from the invoked function. ++ * `pb` must then be freed. + * @param progress_cb_payload Payload for progress callback. + * @return 0 or an error code + */ +diff --git a/src/libgit2/pack-objects.c b/src/libgit2/pack-objects.c +index b2d80cba9..7c331c2d5 100644 +--- a/src/libgit2/pack-objects.c ++++ b/src/libgit2/pack-objects.c +@@ -932,6 +932,9 @@ static int report_delta_progress( + { + int ret; + ++ if (pb->failure) ++ return pb->failure; ++ + if (pb->progress_cb) { + uint64_t current_time = git_time_monotonic(); + uint64_t elapsed = current_time - pb->last_progress_report_time; +@@ -943,8 +946,10 @@ static int report_delta_progress( + GIT_PACKBUILDER_DELTAFICATION, + count, pb->nr_objects, pb->progress_cb_payload); + +- if (ret) ++ if (ret) { ++ pb->failure = ret; + return git_error_set_after_callback(ret); ++ } + } + } + +@@ -976,7 +981,10 @@ static int find_deltas(git_packbuilder *pb, git_pobject **list, + } + + pb->nr_deltified += 1; +- report_delta_progress(pb, pb->nr_deltified, false); ++ if ((error = report_delta_progress(pb, pb->nr_deltified, false)) < 0) { ++ GIT_ASSERT(git_packbuilder__progress_unlock(pb) == 0); ++ goto on_error; ++ } + + po = *list++; + (*list_size)--; +@@ -1124,6 +1132,10 @@ struct thread_params { + size_t depth; + size_t working; + size_t data_ready; ++ ++ /* A pb->progress_cb can stop the packing process by returning an error. ++ When that happens, all threads observe the error and stop voluntarily. */ ++ bool stopped; + }; + + static void *threaded_find_deltas(void *arg) +@@ -1133,7 +1145,12 @@ static void *threaded_find_deltas(void *arg) + while (me->remaining) { + if (find_deltas(me->pb, me->list, &me->remaining, + me->window, me->depth) < 0) { +- ; /* TODO */ ++ me->stopped = true; ++ GIT_ASSERT_WITH_RETVAL(git_packbuilder__progress_lock(me->pb) == 0, NULL); ++ me->working = false; ++ git_cond_signal(&me->pb->progress_cond); ++ GIT_ASSERT_WITH_RETVAL(git_packbuilder__progress_unlock(me->pb) == 0, NULL); ++ return NULL; + } + + GIT_ASSERT_WITH_RETVAL(git_packbuilder__progress_lock(me->pb) == 0, NULL); +@@ -1175,8 +1192,7 @@ static int ll_find_deltas(git_packbuilder *pb, git_pobject **list, + pb->nr_threads = git__online_cpus(); + + if (pb->nr_threads <= 1) { +- find_deltas(pb, list, &list_size, window, depth); +- return 0; ++ return find_deltas(pb, list, &list_size, window, depth); + } + + p = git__mallocarray(pb->nr_threads, sizeof(*p)); +@@ -1195,6 +1211,7 @@ static int ll_find_deltas(git_packbuilder *pb, git_pobject **list, + p[i].depth = depth; + p[i].working = 1; + p[i].data_ready = 0; ++ p[i].stopped = 0; + + /* try to split chunks on "path" boundaries */ + while (sub_size && sub_size < list_size && +@@ -1262,7 +1279,7 @@ static int ll_find_deltas(git_packbuilder *pb, git_pobject **list, + (!victim || victim->remaining < p[i].remaining)) + victim = &p[i]; + +- if (victim) { ++ if (victim && !target->stopped) { + sub_size = victim->remaining / 2; + list = victim->list + victim->list_size - sub_size; + while (sub_size && list[0]->hash && +@@ -1286,7 +1303,7 @@ static int ll_find_deltas(git_packbuilder *pb, git_pobject **list, + } + target->list_size = sub_size; + target->remaining = sub_size; +- target->working = 1; ++ target->working = 1; /* even when target->stopped, so that we don't process this thread again */ + GIT_ASSERT(git_packbuilder__progress_unlock(pb) == 0); + + if (git_mutex_lock(&target->mutex)) { +@@ -1299,7 +1316,7 @@ static int ll_find_deltas(git_packbuilder *pb, git_pobject **list, + git_cond_signal(&target->cond); + git_mutex_unlock(&target->mutex); + +- if (!sub_size) { ++ if (target->stopped || !sub_size) { + git_thread_join(&target->thread, NULL); + git_cond_free(&target->cond); + git_mutex_free(&target->mutex); +@@ -1308,7 +1325,7 @@ static int ll_find_deltas(git_packbuilder *pb, git_pobject **list, + } + + git__free(p); +- return 0; ++ return pb->failure; + } + + #else +@@ -1319,6 +1336,7 @@ int git_packbuilder__prepare(git_packbuilder *pb) + { + git_pobject **delta_list; + size_t i, n = 0; ++ int error; + + if (pb->nr_objects == 0 || pb->done) + return 0; /* nothing to do */ +@@ -1327,8 +1345,10 @@ int git_packbuilder__prepare(git_packbuilder *pb) + * Although we do not report progress during deltafication, we + * at least report that we are in the deltafication stage + */ +- if (pb->progress_cb) +- pb->progress_cb(GIT_PACKBUILDER_DELTAFICATION, 0, pb->nr_objects, pb->progress_cb_payload); ++ if (pb->progress_cb) { ++ if ((error = pb->progress_cb(GIT_PACKBUILDER_DELTAFICATION, 0, pb->nr_objects, pb->progress_cb_payload)) < 0) ++ return git_error_set_after_callback(error); ++ } + + delta_list = git__mallocarray(pb->nr_objects, sizeof(*delta_list)); + GIT_ERROR_CHECK_ALLOC(delta_list); +@@ -1345,31 +1365,33 @@ int git_packbuilder__prepare(git_packbuilder *pb) + + if (n > 1) { + git__tsort((void **)delta_list, n, type_size_sort); +- if (ll_find_deltas(pb, delta_list, n, ++ if ((error = ll_find_deltas(pb, delta_list, n, + GIT_PACK_WINDOW + 1, +- GIT_PACK_DEPTH) < 0) { ++ GIT_PACK_DEPTH)) < 0) { + git__free(delta_list); +- return -1; ++ return error; + } + } + +- report_delta_progress(pb, pb->nr_objects, true); ++ error = report_delta_progress(pb, pb->nr_objects, true); + + pb->done = true; + git__free(delta_list); +- return 0; ++ return error; + } + +-#define PREPARE_PACK if (git_packbuilder__prepare(pb) < 0) { return -1; } ++#define PREPARE_PACK error = git_packbuilder__prepare(pb); if (error < 0) { return error; } + + int git_packbuilder_foreach(git_packbuilder *pb, int (*cb)(void *buf, size_t size, void *payload), void *payload) + { ++ int error; + PREPARE_PACK; + return write_pack(pb, cb, payload); + } + + int git_packbuilder__write_buf(git_str *buf, git_packbuilder *pb) + { ++ int error; + PREPARE_PACK; + + return write_pack(pb, &write_pack_buf, buf); +diff --git a/src/libgit2/pack-objects.h b/src/libgit2/pack-objects.h +index bbc8b9430..380a28ebe 100644 +--- a/src/libgit2/pack-objects.h ++++ b/src/libgit2/pack-objects.h +@@ -100,6 +100,10 @@ struct git_packbuilder { + uint64_t last_progress_report_time; + + bool done; ++ ++ /* A non-zero error code in failure causes all threads to shut themselves ++ down. Some functions will return this error code. */ ++ volatile int failure; + }; + + int git_packbuilder__write_buf(git_str *buf, git_packbuilder *pb); +diff --git a/tests/libgit2/pack/cancel.c b/tests/libgit2/pack/cancel.c +new file mode 100644 +index 000000000..a0aa9716a +--- /dev/null ++++ b/tests/libgit2/pack/cancel.c +@@ -0,0 +1,240 @@ ++#include "clar_libgit2.h" ++#include "futils.h" ++#include "pack.h" ++#include "hash.h" ++#include "iterator.h" ++#include "vector.h" ++#include "posix.h" ++#include "hash.h" ++#include "pack-objects.h" ++ ++static git_repository *_repo; ++static git_revwalk *_revwalker; ++static git_packbuilder *_packbuilder; ++static git_indexer *_indexer; ++static git_vector _commits; ++static int _commits_is_initialized; ++static git_indexer_progress _stats; ++ ++extern bool git_disable_pack_keep_file_checks; ++ ++static void pack_packbuilder_init(const char *sandbox) { ++ _repo = cl_git_sandbox_init(sandbox); ++ /* cl_git_pass(p_chdir(sandbox)); */ ++ cl_git_pass(git_revwalk_new(&_revwalker, _repo)); ++ cl_git_pass(git_packbuilder_new(&_packbuilder, _repo)); ++ cl_git_pass(git_vector_init(&_commits, 0, NULL)); ++ _commits_is_initialized = 1; ++ memset(&_stats, 0, sizeof(_stats)); ++ p_fsync__cnt = 0; ++} ++ ++void test_pack_cancel__initialize(void) ++{ ++ pack_packbuilder_init("small.git"); ++} ++ ++void test_pack_cancel__cleanup(void) ++{ ++ git_oid *o; ++ unsigned int i; ++ ++ cl_git_pass(git_libgit2_opts(GIT_OPT_ENABLE_FSYNC_GITDIR, 0)); ++ cl_git_pass(git_libgit2_opts(GIT_OPT_DISABLE_PACK_KEEP_FILE_CHECKS, false)); ++ ++ if (_commits_is_initialized) { ++ _commits_is_initialized = 0; ++ git_vector_foreach(&_commits, i, o) { ++ git__free(o); ++ } ++ git_vector_free(&_commits); ++ } ++ ++ git_packbuilder_free(_packbuilder); ++ _packbuilder = NULL; ++ ++ git_revwalk_free(_revwalker); ++ _revwalker = NULL; ++ ++ git_indexer_free(_indexer); ++ _indexer = NULL; ++ ++ /* cl_git_pass(p_chdir("..")); */ ++ cl_git_sandbox_cleanup(); ++ _repo = NULL; ++} ++ ++static int seed_packbuilder(void) ++{ ++ int error; ++ git_oid oid, *o; ++ unsigned int i; ++ ++ git_revwalk_sorting(_revwalker, GIT_SORT_TIME); ++ cl_git_pass(git_revwalk_push_ref(_revwalker, "HEAD")); ++ ++ while (git_revwalk_next(&oid, _revwalker) == 0) { ++ o = git__malloc(sizeof(git_oid)); ++ cl_assert(o != NULL); ++ git_oid_cpy(o, &oid); ++ cl_git_pass(git_vector_insert(&_commits, o)); ++ } ++ ++ git_vector_foreach(&_commits, i, o) { ++ if((error = git_packbuilder_insert(_packbuilder, o, NULL)) < 0) ++ return error; ++ } ++ ++ git_vector_foreach(&_commits, i, o) { ++ git_object *obj; ++ cl_git_pass(git_object_lookup(&obj, _repo, o, GIT_OBJECT_COMMIT)); ++ error = git_packbuilder_insert_tree(_packbuilder, ++ git_commit_tree_id((git_commit *)obj)); ++ git_object_free(obj); ++ if (error < 0) ++ return error; ++ } ++ ++ return 0; ++} ++ ++static int fail_stage; ++ ++static int packbuilder_cancel_after_n_calls_cb(int stage, uint32_t current, uint32_t total, void *payload) ++{ ++ ++ /* Force the callback to run again on the next opportunity regardless ++ of how fast we're running. */ ++ _packbuilder->last_progress_report_time = 0; ++ ++ if (stage == fail_stage) { ++ int *calls = (int *)payload; ++ int n = *calls; ++ /* Always decrement, including past zero. This way the error is only ++ triggered once, making sure it is picked up immediately. */ ++ --*calls; ++ if (n == 0) ++ return GIT_EUSER; ++ } ++ ++ return 0; ++} ++ ++static void test_cancel(int n) ++{ ++ ++ int calls_remaining = n; ++ int err; ++ git_buf buf = GIT_BUF_INIT; ++ ++ /* Switch to a small repository, so that `packbuilder_cancel_after_n_calls_cb` ++ can hack the time to call the callback on every opportunity. */ ++ ++ cl_git_pass(git_packbuilder_set_callbacks(_packbuilder, &packbuilder_cancel_after_n_calls_cb, &calls_remaining)); ++ err = seed_packbuilder(); ++ if (!err) ++ err = git_packbuilder_write_buf(&buf, _packbuilder); ++ ++ cl_assert_equal_i(GIT_EUSER, err); ++} ++void test_pack_cancel__cancel_after_add_0(void) ++{ ++ fail_stage = GIT_PACKBUILDER_ADDING_OBJECTS; ++ test_cancel(0); ++} ++ ++void test_pack_cancel__cancel_after_add_1(void) ++{ ++ cl_skip(); ++ fail_stage = GIT_PACKBUILDER_ADDING_OBJECTS; ++ test_cancel(1); ++} ++ ++void test_pack_cancel__cancel_after_delta_0(void) ++{ ++ fail_stage = GIT_PACKBUILDER_DELTAFICATION; ++ test_cancel(0); ++} ++ ++void test_pack_cancel__cancel_after_delta_1(void) ++{ ++ fail_stage = GIT_PACKBUILDER_DELTAFICATION; ++ test_cancel(1); ++} ++ ++void test_pack_cancel__cancel_after_delta_0_threaded(void) ++{ ++#ifdef GIT_THREADS ++ git_packbuilder_set_threads(_packbuilder, 8); ++ fail_stage = GIT_PACKBUILDER_DELTAFICATION; ++ test_cancel(0); ++#else ++ cl_skip(); ++#endif ++} ++ ++void test_pack_cancel__cancel_after_delta_1_threaded(void) ++{ ++#ifdef GIT_THREADS ++ git_packbuilder_set_threads(_packbuilder, 8); ++ fail_stage = GIT_PACKBUILDER_DELTAFICATION; ++ test_cancel(1); ++#else ++ cl_skip(); ++#endif ++} ++ ++static int foreach_cb(void *buf, size_t len, void *payload) ++{ ++ git_indexer *idx = (git_indexer *) payload; ++ cl_git_pass(git_indexer_append(idx, buf, len, &_stats)); ++ return 0; ++} ++ ++void test_pack_cancel__foreach(void) ++{ ++ git_indexer *idx; ++ ++ seed_packbuilder(); ++ ++#ifdef GIT_EXPERIMENTAL_SHA256 ++ cl_git_pass(git_indexer_new(&idx, ".", GIT_OID_SHA1, NULL)); ++#else ++ cl_git_pass(git_indexer_new(&idx, ".", 0, NULL, NULL)); ++#endif ++ ++ cl_git_pass(git_packbuilder_foreach(_packbuilder, foreach_cb, idx)); ++ cl_git_pass(git_indexer_commit(idx, &_stats)); ++ git_indexer_free(idx); ++} ++ ++static int foreach_cancel_cb(void *buf, size_t len, void *payload) ++{ ++ git_indexer *idx = (git_indexer *)payload; ++ cl_git_pass(git_indexer_append(idx, buf, len, &_stats)); ++ return (_stats.total_objects > 2) ? -1111 : 0; ++} ++ ++void test_pack_cancel__foreach_with_cancel(void) ++{ ++ git_indexer *idx; ++ ++ seed_packbuilder(); ++ ++#ifdef GIT_EXPERIMENTAL_SHA256 ++ cl_git_pass(git_indexer_new(&idx, ".", GIT_OID_SHA1, NULL)); ++#else ++ cl_git_pass(git_indexer_new(&idx, ".", 0, NULL, NULL)); ++#endif ++ ++ cl_git_fail_with( ++ git_packbuilder_foreach(_packbuilder, foreach_cancel_cb, idx), -1111); ++ git_indexer_free(idx); ++} ++ ++void test_pack_cancel__keep_file_check(void) ++{ ++ assert(!git_disable_pack_keep_file_checks); ++ cl_git_pass(git_libgit2_opts(GIT_OPT_DISABLE_PACK_KEEP_FILE_CHECKS, true)); ++ assert(git_disable_pack_keep_file_checks); ++} +diff --git a/tests/resources/small.git/HEAD b/tests/resources/small.git/HEAD +new file mode 100644 +index 0000000000000000000000000000000000000000..cb089cd89a7d7686d284d8761201649346b5aa1c +GIT binary patch +literal 23 +ecmXR)O|w!cN=+-)&qz&7Db~+TEG|hc;sO9;xClW2 + +literal 0 +HcmV?d00001 + +diff --git a/tests/resources/small.git/config b/tests/resources/small.git/config +new file mode 100644 +index 0000000000000000000000000000000000000000..07d359d07cf1ed0c0074fdad71ffff5942f0adfa +GIT binary patch +literal 66 +zcmaz}&M!)h<>D+#Eyypk5{uv*03B5png9R* + +literal 0 +HcmV?d00001 + +diff --git a/tests/resources/small.git/description b/tests/resources/small.git/description +new file mode 100644 +index 0000000000000000000000000000000000000000..498b267a8c7812490d6479839c5577eaaec79d62 +GIT binary patch +literal 73 +zcmWH|%S+5nO;IRHEyyp$t+PQ$;d2LNXyJgRZve!Elw`VEGWs$&r??@ +Q$yWgB0LrH#Y0~2Y0PnOK(EtDd + +literal 0 +HcmV?d00001 + +diff --git a/tests/resources/small.git/hooks/applypatch-msg.sample b/tests/resources/small.git/hooks/applypatch-msg.sample +new file mode 100755 +index 0000000000000000000000000000000000000000..dcbf8167fa503f96ff6a39c68409007eadc9b1f3 +GIT binary patch +literal 535 +zcmY+AX;Q;542A#a6e8^~FyI8r&I~hf2QJ{GO6(?HuvEG*+#R{4EI%zhfA8r{j%sh$ +zHE~E-UtQd8{bq4@*S%jq3@bmxwQDXGv#o!N`o3AHMw3xD)hy0#>&E&zzl%vRffomqo=v6>_2NRa#TwDdYvTVQyueO*15Nlo%=#DXgC0bhF3vTa`LQGaO9;jeD$OP?~ +za$G4Q{z+Q_{5V?5h;a-noM$P{<>Q~j4o7u%#P6^o^16{y*jU=-K8GYD_dUtdj4FSx +zSC0C!DvAnv%S!4dgk +XB^)11aoGMJPCqWs%IS0YSv(eBT&%T6 + +literal 0 +HcmV?d00001 + +diff --git a/tests/resources/small.git/hooks/commit-msg.sample b/tests/resources/small.git/hooks/commit-msg.sample +new file mode 100755 +index 0000000000000000000000000000000000000000..f3780f92349638ebe32f6baf24c7c3027675d7c9 +GIT binary patch +literal 953 +zcmaJy@-{3h^^Cx;#d0zEA@DDc$nY4ez&|=%jTg@_HU*ub=!!y$xW09TSjlj +z(`I@QCsM`!9&80$I98wsQ8yK#)Orb<8re8FjkKh630D$QUDwi~(gkX=RunYm$rDjk +zlp%RUSnzA#6yjdG5?T?2DcYKp+v_lts0ljn&bh3J0bD5@N@1UKZ190O6ZeWr-BuZ^ +zWRebCX%(%=Xoj#(xYk1Cjtr!=tyBesf@m6}8zY6Ijbz9i9ziI_jG9MvR +zDH*e>^ga9IR?2wrSrAVm;eButj4Y>7(E2?b~jsu>& +zRKCJ7bp#19sqYh627wD%D9R$8=Ml$TNlumDypl~$jBu*G>5fIR^FB0h0Ex&TGZNr> +zL5hs1_K>taRb!|ThN9ns7^@4MXKP+6aGI_UK)T-M#rcP$;kN(Vcf#P)+5GzWa{l@J +z>-E{`$1iiNVYxq27}j;uo%;)r3kJI2xCFF~Ux;$Q%) +wjbk6JlDCM`jU&P+UVOvg`|iYl<7~9k>HHB4I;pdlQ=I-^$DrHaN$@lH1?P!0U;qFB + +literal 0 +HcmV?d00001 + +diff --git a/tests/resources/small.git/hooks/fsmonitor-watchman.sample b/tests/resources/small.git/hooks/fsmonitor-watchman.sample +new file mode 100755 +index 0000000000000000000000000000000000000000..41184ebc318c159f51cd1ebe2290559805df89d8 +GIT binary patch +literal 4777 +zcmbtYYi}F368$Xwipg4lq(BeHMvzvH-4;n7DGJBPqq#tw3aed8+IU5-m)yvL>;Cqh +z8FFRGj$`9CA8aoJ?j^$%==FV``-=rhLcPW`McSytRm~mEO7_&_cAVZrf1fFy*ha@8oe%*-aBYE +zcjzZg>LOkgxuUr-XJnHyD;zmPnRaSc#!k_P*d_BttRdc+J6G7za5#+^Y1nkc2Oowk`ya47uUR3Feu?B(w;S{(VYzxh}q-=#zP@uxSx{wbyPUMFU;K(06)$o{07&3yI?q{GqMcQ1c_^M<0< +zF4acAV)Il-V(rCTC1(;bsZ*}bl8dmejAk~yb`B}!^0;g^(o9kGUfZfDOvyp@x4OQt +zSgWh6T|3eq;9MFs8-#z+FDM1h(IjRUP|``PxupgJ7CUHOH90gbgl^2~97`?_X{P)) +zB*$r1cDlF-%azKND}?Gv`2K8-9v5e`gQoft=j?T<&a13c^!wY_$D`5z-X1g?ty&6- +zQN50{8?bUk9AI->^W@~~nkOghHIC2YN+AXkLQG_2-{Pq3%{`3KUMeG$iIn%%^6*NYb +zn|_BdV#C)n4565VccX;uT8&z3vSi!HXGbUj2B!R +zdz~&#fk#L-&k$fLwo$4?>12g@AXOKFekuo#6EHB%gmpD?1eyh%N8s{2wGoTu +z*@6cEZ^ZW!FAF_|JL`NkV7k}0ow|-2jHwbgH0;c@Dq*o?@&c*HnGdyx6^su8Qk%2{ +z*ye(dxO*6-&>qn1+zw}tc6;=sOX{4WB=VqjTS^))y1jlX2Q;=e!qMmFA5lC$#;BxC +z=Y%tRpWxb+_uQAvAw7Q{HGV#R$xb&udLCzZ+HN?kTyB};1EJ8UlQ5!>5eGW@)RX0n +zkjj>EF!3=0Gl^8dzv$B^NMGRxJoqN4A`xq-@wCbrx*u2NmIJ1xZ%H +zh;{|4T3(!E9sY#Ni(wUJYs1MmIc9bl)(4Nl3_wD_BWB>i<1S(LX7m*{Q7PU$muMS* +zM!%0EZx-Vw=Zey;erC?SNxF;pY@^A%-krqzfLV2meBp1vWdyArFYn`DD19T)Hw(?n +z)}{NP(Lk(o*?gl#B@pP7^*r|=;PIDT4|F#{2Hzh-AL0Rv$6uT;n|WzE4=slK?on@(fZeGhRgQCu56qB +z{+n81Az96qnQjMY*-*r-KV*7;Z#4QuJRJJV$M^KdldiMhj?ImK6~FvwJ*L5a){QoM=L5TYHkGO1$UrO3`a>{?Opw|b +zG(#59NQ#jFL9v~vgOVkM@^^(^A}onOE))yWEwhIlk&{ZyseZ^O0b=w8&O=BK{k<5B +k^Q-B@eG}LeHrquz%(SVEp_N)VhYZikCW__82JXfD17`J9Qvd(} + +literal 0 +HcmV?d00001 + +diff --git a/tests/resources/small.git/hooks/pre-applypatch.sample b/tests/resources/small.git/hooks/pre-applypatch.sample +new file mode 100755 +index 0000000000000000000000000000000000000000..625837e25f91421b8809a097f4a3103dd387ef31 +GIT binary patch +literal 481 +zcmY+ATTa6;5Jms9iouO45IBJXEg&Jm9@v1LPHMM_ZR|;#6tQh$71hSXq*MxP;V& +zj0cY7SCL=x4`a46sF)C>94Gk%=3q$W2s;j6iHtB2$R0%gix4oK@&T~=ALd_o*CKxt +I-`Pv{1Bpzc>;M1& + +literal 0 +HcmV?d00001 + +diff --git a/tests/resources/small.git/hooks/pre-commit.sample b/tests/resources/small.git/hooks/pre-commit.sample +new file mode 100755 +index 0000000000000000000000000000000000000000..10b39b2e26981b8f87ea424e735ef87359066dbb +GIT binary patch +literal 1706 +zcmZuxU2ohr5PY_N#pZ0-F<{-v&v-X^RA+u>k}E$4d&uD7=g_fA8+pNNV=4s0|iD3p<=DTXClTS +zXV23tJ;ECmN@M0j@zUAKEYW@3bv!SeYZ8ZH`YQNTApFVNc;F|9r5p4TqGs=>8E?6y +zi|gY{iM#PG1nL?UE9YCnWTk72kgZPG*Usqw!~Qd3c?~@w2?%eg@~)+VlSs6N5Yf2^ +zz;owF#K#r^&KMq1A`oqVGFpD&-!Pv|Rc +zO3KSqA@h9nSc%bm`0)Amk6*J}@14J*1-219l%%7D!Pl}UK>|lVi0Dfgu2jN3WC!uL +z0ej??b2iSehVgdnWHmZV4kUo*QL#aiIp}U=9x)IXk}JJ7VQ;CI9Rtn5e0VcjbYcVt+`x5D+svCGD;Z5hm*E$jSEQZ%SQ(}oLgslTvrKK@9Qf#b!hajVFnp9@oIix;NcI9Wk +xjnh0ya!AWet{I7YpD;y6HXyzI*lfSvH=o6*7mJZPkuaYpm>vzZ`wyGEBtOQPo|pgt + +literal 0 +HcmV?d00001 + +diff --git a/tests/resources/small.git/hooks/pre-push.sample b/tests/resources/small.git/hooks/pre-push.sample +new file mode 100755 +index 0000000000000000000000000000000000000000..02cbd80c287f959fe33975bb66c56293e3f5b396 +GIT binary patch +literal 1431 +zcmaJ>U60!~5PUX&#a1@z9B{IIZkjLT0t5kq9#8~D(I5{+8&J~9;#ndUk~-ZT`r|uG +z$#K$$J{TsKs*LP1}9!GoZ@4I4myMMG_di|of +z%?llx{O8TS-#^;(OioEmPy%kwWQBA1OMzV{hsQ8XFzS1k!~YQoLa5 +zhtP1fA$q6VmMbbAC_9)4I628k*O5J$NR19uHe4QYDK<==I~SQk)Nu%xQ~KH +z53w=!ke(FGb_PpnZfd*+hnXDTn;2*`u^~;?+5C~cn?bRka7NR%06%e6O91{MAgN6J +zmlO8{Biw4&wr&&(z4p3eln`E}XR9m9bNYZ7Ibrg(4yZIXrfgD7N*AFD7L3YSM#j}% +zo__rOS5fr;@8UM<6cl+cv_$YB$PQ&9dv($eM*))g!_cu!QcSh-mqE9i#QDZT)=o#` +z?8!RtE?w6p?GkGZ-6yt_p~5~4ecu|Sf^)6096%h*q-eNiEA1;Xwg)p~Q&iGSG7-IQ +z9aII&`ps$WOojFA`*bjGkFk|E@sHHuD}W^d`7YJ3YE^zrQnqR +zGoq?;YGKe)93o|_=^f%3U1KYZGPOXRRxK7w`UUbMMa3<86OmVH!EKP$8RCrn9mWX+ +zC?9yF!fRVLmud3hF<}x;;sR}f(*r}6Gap3fR6zLHR~kbMgD{98N`L+r&?3p~*0+FX +zcAL%j=(SO}xTJUTvA`&Lf`2mv4koPG9&|;2+68$XxiXKL@ma;l5d2^5Ba_rPh_DHI-u1#&_upttZXp;no03$20|NFiM +zK#D#xQ>!Z3JkX8T-LDVm!B5j7y_{;JDmmTTef+K1oIiPzeEr+Ai*<2PUgnG4^ZB>p +z_fkAvoR1emuf~ri^K$-px=4#D-vY9w& +z`bCv#2zVn=YnJyeNey(Y +zRh`9vtLw~A+5zsjp|W0Nsa|29Rm!B>OoG5a+vi;ari8O>KkU!KAWg_fa3btK2x*_@ +z0bEc7J;Ubghm}n9bOi(Sv_B66nQ7U)J7f0fO}8Wuf*uorcIgEG +zOHc|-V6+HlRhOP}?Cn?@5iwSl43abmBA^2lyL$+cpabCGVES+v^j^FO_}?FIp%En%Ll?Z*7*}TwrZyg5OSZ9rY-`aU~Mc-jjv{Ll)FLMgtB4ujktfQ`Xhqrka +zT=P!A;9w^;Z?PqpLwOLu=cj3L>TdUKw2;DMu)`oVkj}#bcDx4tYg=j%D`+i{W~fVM +zVmZ>W9VMyin9c-0KzI_;iZ-g|OyzuG`Yq%(%dvl;ifnVr0;jWE&S`z|rQu=!yHBBO +zx`OJ;oOQ(KKM<$(bC38o>pD0%|HA(E0TRw7qj$fJ_pRN+7Nm>dSC(gLg{(`t+5Z=?o+}wXU4tHy+&%F&aRhFebeEhR2R5|$#Ycbp^w@t +zTl%=f1t=w+WpJzF<|CE@?SCNAz)%9?w33lQ8vrHJqPfH9@}qs*QXOG71W=ylx;wOB +zcx!Bj^)Yy6WX$a^vBkBJ5CobqlaDx_B0c<3b+8)f84LCrt;e;qxc+7>VbwVK{skNv!wvBiTa^9Iu +zkwP;VK)jH$WJ{`MRwAA9fal!y0dtV;FWg8PTkWU>CwnqD>1ZX2B@;$DlX%C5MI+}{ +z9xQVnffR*~v2KAUj*hCdgul~`bk#mk`o>zk9)<2Uc8?hUZAEvd!`9em)~$Z)zev>w^8 +zyAgCP_$&Y)7HSQ84`xG}OeTavaEswwF|8Xpi5iZzZa@hCiv(J-%bfFC&)HLlO+Rhw +zG6g?9eL5&A!SuJnQ6}LxG%tU+@vZ`i+!+Rz6iYvsTdhnPo7lW{m-}{hya@viX4)XZ +zngaw+j;gloB#|UwI@8sOmQpc`h+bicQJnQIB5eifIMQNgD2+oai33m!34~xU|0Azj +zhu$8z+T5^;Pxx@d{N)pzOJLSa^e;aDf$W%N5XcOf!mGC9l9j$Ev2h6N+6ZQC+CJzl +zaM7?S!SrFLS2DASjj(h6y1WN3N?|bmqmyzm!&nLoE|`rKBOc_yDF$a#FsUn!IQf(t +zdC&Us(kQz*7mvH^j*^MC@>wTDb}g%~sx*ng#>{@lR=XG-Z5_ +z#<9*Oh0joMzt;nS)ObAp)347`D=}r-;nV!TbIq&xrGRGsF6fZg+!VkfUei@_&l-M& +zPqQ+Dw)RV}+)I8RuqAxa`Pv8e&!_gXS=e2-un>=Ktn}-;%lLZxaVn?Q>yZCb2R3Wk +z77zr%;Rq&h|2ncqyKYmFI0148JVY7Q$V5p=dWj+Qqpu%i|xp2C=WaOb2Wudn^h0EcD%$p9YVU1fnoRV9`(cy(vv6K>FXS!2jY>1GnU--7)4usH&K +zao*&P^@9~YmUe|ZdLW@C>H;!*Vt3>Nw4M*;=?j(TBD#O@XCv0|MEhA;z}kTFRv@`tPHhp=&Yh +zg%Zhg4i7o_k{a5i&f5;tZ==%}^Sn4aD_6%qs_XAuJt&EumdH4Yu`UjT<-+XHTuHss+b +YOmM2;hq8Egm*4=7_P9T{21QBYH*F=mfB*mh + +literal 0 +HcmV?d00001 + +diff --git a/tests/resources/small.git/hooks/prepare-commit-msg.sample b/tests/resources/small.git/hooks/prepare-commit-msg.sample +new file mode 100755 +index 0000000000000000000000000000000000000000..b1970da1b6d3f42f00069fd17c325de72cda812e +GIT binary patch +literal 1702 +zcmb_cTW{Mo6n>t6#i?x6xmZ$SFLf{QfG*3r0L?Pg?px55l8$UTGO3bO;spKi{V3XX +z))weX0X>M9bNMcZ-6yG%>(n}JI2|25dr}WZBP@ih?JX^+@ +zu#5O48P>yRX(mfDIhYP)doc1&TADZa@ZGpusJ$6G+e$ZMcmC +zoOosDQPS}l{H?YPsq(4;0SGkATa9eeqAaDcjq8n2wALbFwU@2i@FAaRV!=uw-nwx1gKn2SvY +z>Ff>;2sg!+Hxfkwv1lsiii=p6WenF=5)6LZcQaZ=aS_}+-4Y&?!@HWh|<^gJ21!|T@+%On#w6azxPHV}XsRbe*w +zR_TZ2XEsQa1lPK~biYqg@0-RW@5J1@=<87cFzEUABdCoFH2CZo?}l(Z*!OFqUxo>K +z_d`l#4d9|H6;VPT{X?^{VJ>oL|D7K{BJwwqB>`YcPoGk+9hbvHnoQ{EM|kPgD_`wk +zKm4#2xu;-y`RAm!=L_BnLvJ8$AZm8@?)v<%vwvsw8AF2x6!mTT;c72A_~U9nIq0ST +zv)N0!I!^1p=g8-RQfx5)E_Mb_4I2vtQpI30XZ&t-9h5!Hn + +literal 0 +HcmV?d00001 + +diff --git a/tests/resources/small.git/hooks/push-to-checkout.sample b/tests/resources/small.git/hooks/push-to-checkout.sample +new file mode 100755 +index 0000000000000000000000000000000000000000..a80611e18896f212c390d845e49a3f6d5693b41d +GIT binary patch +literal 2840 +zcmai0U31$u5PXh)#YOS7cE^-rw@uolNhe9&aUS|HtvhX>G$45tVUYj>fRdF?|9kfU +zNR~aG=E)WbEbeyq7JTw}ZuHIE2kUtL<AoeCNptd-NM1aZLhESzC;I`+Ns +zfmNNjdAp^W8#Q*}l>CT7RB9F5(BbI8ly2l~+E};JW|>&d1)=epZ-8vm8ppkbEVn#R +zt30a5A-c(YQR8eM5%;|UAnO>rt!&@x@G@yp+92%w-}%(5P_+P&Wf_zb$f-Qrl5(7z +z2ah(bkE;!DK(&aAMuQ%1TS>ai?wSXCOCSj=_}8x4IbCx^$}9q)whwv)SBt| +zg#MX4;;Oau`m=MI9(^&zPbueY@~>3*ixX%mvR5m_1&nAg@ZKvY1E$O}&EtLiG;mhV +z1xhMIm~fGjmf_#{62f`y;09?I7M1W2tWQvz<}i9lR>OpQyUJi45_&*pQus&EkwY<> +zI|ZAx=*3i9a-)g)hXkvO7>UJ5MNgL(Z+-wpXVcgbSgpmFmbf1~DPA(OVGI&FNLeIE +zNH!_aiH$vsif$_j7=T2{cS(!DOI`~bn@)vSd-0d7xL=DF;UNP|tW}4ih>DvHtu9tY_pbJ6x(6E*hxgC +zzNDao%qlr-IE%YGbS4hF!n!on7#W3$bX-_hbZAaws^nHu#)Dx=WzdbJ>AKzAy@T$x +zSWE^x9+|TEHVEPyaPYa0DOChp?AeHSBBDbZNokQpAY{lE!7geZI=jV)G^2@l)&91Zb1+`T+oq9wWF +zRV~kGTGce0O~p^6mj{kT5kL(pv>r;Lvd7VDX*P>A^Th`$3cWO0L81p4Ysdo3ZP1(SrR-peEdTo;-@bkB((G +zPHYQXUL!@Q$e(OQ;R9r%@Afz+50I7>*^^c&&|E*r-jN)LH=pM4AqMwWxSv|nqjddE +Z4{_hwv8!W(T +zYw`X3V>TCdnSD1ru8&`j=2DIPbCT@SnIgUw>$+lEYP}+x8(BMYnr=iT3*ndq)xzaV +z>I+qjv}vC#8_9M+b1p#uNS0M0)q

8!3p_LRQ0MA3M`!2foxzRUjbFY@}O~(ki=S +zqscnq8cU*dY)D$$cqE}n)V0yIk>CNKHCrndOtSP*HbOb;nbwAHSb;R+gs^?^Dve%) +zoW}t(*D}$>O3ab0TS^-;J|u&sb-PkZzo#kn*#xYt(;FGuwzSb^g&RDiGcOz9TB;Hu`nJh)$W=C=XCSm2AY=$w3G3P-V#Oo+N*;#2 +z4ijJ-pBZ=;T(RTgp_HYrD!uW-dTMfkuqY5jwOy)~gM;#=P^i{!l7`pXTS^s(&^{RU +zydaw}OpS#^D1cXM8?FW+fh`t7D(g;yr6|}fdaNtZBx3hlK~IpkTu3!Qq%R+zAo#t}Bs8^3$vHD+-TGT@`F>H1Cc#WAVW;&$S6%fE2d6@kLS0g&ihIM{}0z +z8#XhD>b>3{(BH|Px7}&lJ4%y1v(CihZJx@8MPoGdl*BJGD;usf*iS7%;{Joe; +zNFuBa>*~o&qETDPo~u&~$FxE1xb^x&(CbE`Y3GfsibL2rl+L;>P6j&Y3U>K$mkp*6 +zd`Q{<^+^&;GskGjwD-%!boR&i-TCA9UOR|@=GYb5x#+dhd7fkaVIR^pol`Mv+rUbmZ43dVL6^S7g3{NsPiG$iy$5EDB% +z6KIgnb$H(n&t3e4E6d4V7w^B?JS}JkG)PM6+X3Co`SQs($O*AA+MG~{S7RJ=cy-l& +z>~%3y`tjfx2>uOutB_^s +ziwG=e=ch|FQ0IkN91US7rhdQkXhwwt$gU0WEVDjo=IPb+?6PC=s8}J*ua(Ms))`UL +fi$|vMHn?H_tSE3ettp-hLlsZCxaLX8(nU;bVRB;Ce6@s#eu2|WvLz>- +zvy(&>Gyfp@+BtKnpqWkKi^+v{4jn_pNw_zeuxETifiGO|)w}OANj2n2D^K=o3j6P6uOL70#cbA{uzWXDlk1wr9GV1X(2W{RuTvjXV +zCmd8u +zH%V`94=q3)Dk)PHNrnFC(T1)Om6f{Usj;u1R->&XoCYVK2V3ZlgZuF?N}1+33OER*x +z*9Z=L=zI8CN>A_^jYjt0F$psO$sL=38q5q|SG)qCN6{^>RFh5E&l5GZ$pEahnF&d+ +z5c>64t}uJPkf~_!VUj#&N%nC-gUMj%=@B=!V>&}xtj2%@-mOm#rQUSJ3(ccmc+fza +znZ#uxF>N?QN5UrIEd!5RgHEfW#;(nKYF+D<*rdshJ$X-z2OZ2X;)nn@KSVdVhaA?}@3;6gZxb4v +zozoWSr{{+!h}zGpumG3H`=AvWpm^9kW;J$Jp^Xl*?8ckr`fqN%c|Z;VC0|cM4vSrk +zH_O8Yvh85nvJp^;``wo8=z0f`FWg?`>gO#y1hjX1{}rTlg9rwIKia8eyGexA3GnuR +z`Rg~XZoW;0pA)vI8=p5!+6sIn#C^FCvR>ffv39h6SCNi9v);%WD;WZ`of_MgwyRWy +z-yY%n*Y>X89W-v4`Ff%bx$Vkn}$!Ay}rnY6F$m-Kg*KD_+;Lx#g4|^&N +I02NaX#p`nv=Kufz + +literal 0 +HcmV?d00001 + +diff --git a/tests/resources/small.git/objects/af/5626b4a114abcb82d63db7c8082c3c4756e51b b/tests/resources/small.git/objects/af/5626b4a114abcb82d63db7c8082c3c4756e51b +new file mode 100644 +index 0000000000000000000000000000000000000000..822bc151862ec3763cf2d3fa2372b93bbd3a4b65 +GIT binary patch +literal 30 +mcmb>0i}&W3IZ_@1U=^!a~EV1casc=c+{&un1qQN*i9hD|0|m(2n|iwp*q%W +z%N;b$hu%cM`$TMo*~EnC1BFP&Pfj~;jZVKXQ96s_PhV<-XAROi+@-v8dBLUa`!;GB +k^iXlEv8$>R)1G>9th&t3j;s7J{?^9n|7U^`%mXoWC24Q^m!3%@{ + +literal 0 +HcmV?d00001 + From 976f539f7dd2ee6345b605def105883d16b32c60 Mon Sep 17 00:00:00 2001 From: Robert Hensing Date: Wed, 28 Aug 2024 02:35:46 +0200 Subject: [PATCH 7/8] Make Repo::flush interruptible --- src/libfetchers/git-utils.cc | 115 ++++++++++++++++++++++++++--------- 1 file changed, 85 insertions(+), 30 deletions(-) diff --git a/src/libfetchers/git-utils.cc b/src/libfetchers/git-utils.cc index 70391a287..265eb35d2 100644 --- a/src/libfetchers/git-utils.cc +++ b/src/libfetchers/git-utils.cc @@ -166,6 +166,45 @@ static Object peelToTreeOrBlob(git_object * obj) return peelObject(obj, GIT_OBJECT_TREE); } +struct PackBuilderContext { + std::exception_ptr exception; + + void handleException(const char * activity, int errCode) + { + switch (errCode) { + case GIT_OK: + break; + case GIT_EUSER: + if (!exception) + panic("PackBuilderContext::handleException: user error, but exception was not set"); + + std::rethrow_exception(exception); + default: + throw Error("%s: %i, %s", Uncolored(activity), errCode, git_error_last()->message); + } + } +}; + +extern "C" { + +/** + * A `git_packbuilder_progress` implementation that aborts the pack building if needed. + */ +static int packBuilderProgressCheckInterrupt(int stage, uint32_t current, uint32_t total, void *payload) +{ + PackBuilderContext & args = * (PackBuilderContext *) payload; + try { + checkInterrupt(); + return GIT_OK; + } catch (const std::exception & e) { + args.exception = std::current_exception(); + return GIT_EUSER; + } +}; +static git_packbuilder_progress PACKBUILDER_PROGRESS_CHECK_INTERRUPT = &packBuilderProgressCheckInterrupt; + +} // extern "C" + struct GitRepoImpl : GitRepo, std::enable_shared_from_this { /** Location of the repository on disk. */ @@ -213,42 +252,58 @@ struct GitRepoImpl : GitRepo, std::enable_shared_from_this } void flush() override { + checkInterrupt(); + git_buf buf = GIT_BUF_INIT; - try { - PackBuilder packBuilder; - git_packbuilder_new(Setter(packBuilder), *this); - checkInterrupt(); - git_mempack_write_thin_pack(mempack_backend, packBuilder.get()); - checkInterrupt(); - // TODO make git_packbuilder_write_buf() interruptible - git_packbuilder_write_buf(&buf, packBuilder.get()); - checkInterrupt(); + Finally _disposeBuf { [&] { git_buf_dispose(&buf); } }; + PackBuilder packBuilder; + PackBuilderContext packBuilderContext; + git_packbuilder_new(Setter(packBuilder), *this); + git_packbuilder_set_callbacks(packBuilder.get(), PACKBUILDER_PROGRESS_CHECK_INTERRUPT, &packBuilderContext); + git_packbuilder_set_threads(packBuilder.get(), 0 /* autodetect */); - std::string repo_path = std::string(git_repository_path(repo.get())); - while (!repo_path.empty() && repo_path.back() == '/') - repo_path.pop_back(); - std::string pack_dir_path = repo_path + "/objects/pack"; + packBuilderContext.handleException( + "preparing packfile", + git_mempack_write_thin_pack(mempack_backend, packBuilder.get()) + ); + checkInterrupt(); + packBuilderContext.handleException( + "writing packfile", + git_packbuilder_write_buf(&buf, packBuilder.get()) + ); + checkInterrupt(); - // TODO: could the indexing be done in a separate thread? - Indexer indexer; - git_indexer_progress stats; - if (git_indexer_new(Setter(indexer), pack_dir_path.c_str(), 0, nullptr, nullptr)) - throw Error("creating git packfile indexer: %s", git_error_last()->message); - // TODO: feed buf in (fairly large) chunk to make this interruptible - if (git_indexer_append(indexer.get(), buf.ptr, buf.size, &stats)) + std::string repo_path = std::string(git_repository_path(repo.get())); + while (!repo_path.empty() && repo_path.back() == '/') + repo_path.pop_back(); + std::string pack_dir_path = repo_path + "/objects/pack"; + + // TODO (performance): could the indexing be done in a separate thread? + // we'd need a more streaming variation of + // git_packbuilder_write_buf, or incur the cost of + // copying parts of the buffer to a separate thread. + // (synchronously on the git_packbuilder_write_buf thread) + Indexer indexer; + git_indexer_progress stats; + if (git_indexer_new(Setter(indexer), pack_dir_path.c_str(), 0, nullptr, nullptr)) + throw Error("creating git packfile indexer: %s", git_error_last()->message); + + // TODO: provide index callback for checkInterrupt() termination + // though this is about an order of magnitude faster than the packbuilder + // expect up to 1 sec latency due to uninterruptible git_indexer_append. + constexpr size_t chunkSize = 128 * 1024; + for (size_t offset = 0; offset < buf.size; offset += chunkSize) { + if (git_indexer_append(indexer.get(), buf.ptr + offset, std::min(chunkSize, buf.size - offset), &stats)) throw Error("appending to git packfile index: %s", git_error_last()->message); checkInterrupt(); - if (git_indexer_commit(indexer.get(), &stats)) - throw Error("committing git packfile index: %s", git_error_last()->message); - - if (git_mempack_reset(mempack_backend)) - throw Error("resetting git mempack backend: %s", git_error_last()->message); - - git_buf_dispose(&buf); - } catch (...) { - git_buf_dispose(&buf); - throw; } + + if (git_indexer_commit(indexer.get(), &stats)) + throw Error("committing git packfile index: %s", git_error_last()->message); + + if (git_mempack_reset(mempack_backend)) + throw Error("resetting git mempack backend: %s", git_error_last()->message); + checkInterrupt(); } From 459d02672c5ea9c6b909c0ace6b11141af79421e Mon Sep 17 00:00:00 2001 From: Robert Hensing Date: Wed, 28 Aug 2024 11:13:03 +0200 Subject: [PATCH 8/8] fix Windows build --- packaging/dependencies.nix | 19 ++++++++++++------- 1 file changed, 12 insertions(+), 7 deletions(-) diff --git a/packaging/dependencies.nix b/packaging/dependencies.nix index 0182f29c0..e5f4c0f91 100644 --- a/packaging/dependencies.nix +++ b/packaging/dependencies.nix @@ -116,20 +116,25 @@ scope: { cmakeFlags = attrs.cmakeFlags or [] ++ [ "-DUSE_SSH=exec" ]; nativeBuildInputs = attrs.nativeBuildInputs or [] - ++ [ + # gitMinimal does not build on Windows. See packbuilder patch. + ++ lib.optionals (!stdenv.hostPlatform.isWindows) [ # Needed for `git apply`; see `prePatch` pkgs.buildPackages.gitMinimal ]; # Only `git apply` can handle git binary patches - prePatch = '' - patch() { - git apply - } - ''; + prePatch = attrs.prePatch or "" + + lib.optionalString (!stdenv.hostPlatform.isWindows) '' + patch() { + git apply + } + ''; patches = attrs.patches or [] ++ [ ./patches/libgit2-mempack-thin-packfile.patch - + ] + # gitMinimal does not build on Windows, but fortunately this patch only + # impacts interruptibility + ++ lib.optionals (!stdenv.hostPlatform.isWindows) [ # binary patch; see `prePatch` ./patches/libgit2-packbuilder-callback-interruptible.patch ];