Message ID | 20240222175033.1489723-3-calvinwan@google.com (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
Series | None | expand |
Hi Calvin On 22/02/2024 17:50, Calvin Wan wrote: > This commit contains: > - Makefile rules for git-std-lib.a > - code and Makefile rules for git-stub-lib.a > - description and rationale of the above in Documentation/ We tend to avoid lists like this in our commit messages. Starting with the motivation for adding git-std-lib would be more helpful to the reader I think. > Quoting from documentation introduced in this commit: > > The Git Standard Library intends to serve as the foundational library > and root dependency that other libraries in Git will be built off > of. That is to say, suppose we have libraries X and Y; a user that > wants to use X and Y would need to include X, Y, and this Git Standard > Library. > > Code demonstrating the use of git-std-lib.a and git-stub-lib.a will be > in a subsequent commit. > > Signed-off-by: Calvin Wan <calvinwan@google.com> > Helped-by: Phillip Wood <phillip.wood123@gmail.com> > Signed-off-by: Jonathan Tan <jonathantanmy@google.com> I think virtually all the changes since the last version apart from rewording the documentation come from my fixup at [1] I'm happy to offer my Signed-off-by: for those. I cannot offer a review of the changes from that fixup though I'm still happy with the approach. I do however think I should have included git-compat-util.h in the stub implementations and used BUG() instead of assert(). I've left some comments on the documentation below. [1] https://github.com/phillipwood/git/commit/0f8393d2189a4c73d3f00f5ae74d3972677309d0 > diff --git a/Documentation/technical/git-std-lib.txt b/Documentation/technical/git-std-lib.txt > new file mode 100644 > index 0000000000..3d9aa121ac > --- /dev/null > +++ b/Documentation/technical/git-std-lib.txt > @@ -0,0 +1,170 @@ > += Git Standard Library > + > +The Git Standard Library intends to serve as the foundational library > +and root dependency that other libraries in Git will be built off of. > +That is to say, suppose we have libraries X and Y; a user that wants to > +use X and Y would need to include X, Y, and this Git Standard Library. > +This does not mean that the Git Standard Library will be the only > +possible root dependency in the future, but rather the most significant > +and widely used one. Git itself is also built off of the Git Standard > +Library. > + > +== Dependency graph in libified Git > + > +Before the introduction of the Git Standard Library, all objects defined > +in the Git library are compiled and archived into a singular file, s/singular/single/ there are some instances of "singular" later on as well all of which would be better as "single" I think. > +libgit.a, which is then linked against by common-main.o with other > +external dependencies and turned into the Git executable. I found this description a bit confusing. As I understand it to build git we link git.o against common-main.o, libgit.a, xdiff/lib.a, reftable/libreftable.a and libpcre etc. > In other > +words, the Git executable has dependencies on libgit.a and a couple of > +external libraries. The libfication of Git slightly alters this build > +flow by separating out libgit.a into libgit.a and git-std-lib.a. > + > +With our current method of building Git, we can imagine the dependency s/imagine/visualize/ > +graph as such: > + > + Git > + /\ > + / \ > + / \ > + libgit.a ext deps > + > +We want to separate out potential libraries from libgit.a and have > +libgit.a depend on them, which would possibly look like: > + > + Git > + /\ > + / \ > + / \ > + libgit.a ext deps > + /\ > + / \ > + / \ > +object-store.a (other lib) > + | / > + | / > + | / > + | / > + | / > + | / > + | / > +git-std-lib.a > + > +Instead of containing all objects in Git, libgit.a would contain objects > +that are not built by libraries it links against. Consequently, if s/by libraries/by the libraries/ > +someone wanted a custom build of Git with a custom implementation of the > +object store, they would only have to swap out object-store.a rather > +than do a hard fork of Git. > + > +== Rationale behind Git Standard Library > + > +The rationale behind the selected object files in the Git Standard > +Library is the result of two observations within the Git > +codebase: > + 1. every file includes git-compat-util.h which defines functions > + in a couple of different files > + 2. wrapper.c + usage.c have difficult-to-separate circular s/+/and/ > + dependencies with each other and other files. > + > +=== Ubiquity of git-compat-util.h and circular dependencies > + > +Every file in the Git codebase includes git-compat-util.h. It serves as > +"a compatibility aid that isolates the knowledge of platform specific > +inclusion order and what feature macros to define before including which > +system header" (Junio[1]). Since every file includes git-compat-util.h, > +and git-compat-util.h includes wrapper.h and usage.h, it would make > +sense for wrapper.c and usage.c to be a part of the root library. They > +have difficult to separate circular dependencies with each other so it > +would impractical for them to be independent libraries. Wrapper.c has > +dependencies on parse.c, abspath.c, strbuf.c, which in turn also have > +dependencies on usage.c and wrapper.c - more circular dependencies. > + > +=== Tradeoff between swappability and refactoring > + > +From the above dependency graph, we can see that git-std-lib.a could be > +many smaller libraries rather than a singular library. So why choose a > +singular library when multiple libraries can be individually easier to > +swap and are more modular? A singular library requires less work to > +separate out circular dependencies within itself so it becomes a > +tradeoff question between work and reward. While there may be a point in > +the future where a file like usage.c would want its own library so that > +someone can have custom die() or error(), the work required to refactor > +out the circular dependencies in some files would be enormous due to > +their ubiquity so therefore I believe it is not worth the tradeoff I'm not sure if we want to use the first person in our technical documentation, unlike the cover letter to a patch series it is not immediately obvious to the reader who "I" is. This applies the passages in the first person below as well. > +currently. Additionally, we can in the future choose to do this refactor > +and change the API for the library if there becomes enough of a reason > +to do so (remember we are avoiding promising stability of the interfaces > +of those libraries). > + > +=== Reuse of compatibility functions in git-compat-util.h > + > +Most functions defined in git-compat-util.h are implemented in compat/ > +and have dependencies limited to strbuf.h and wrapper.h so they can be > +easily included in git-std-lib.a, which as a root dependency means that > +higher level libraries do not have to worry about compatibility files in > +compat/. The rest of the functions defined in git-compat-util.h are > +implemented in top level files and are hidden behind > +an #ifdef if their implementation is not in git-std-lib.a. I think the reference to #ifdef is out of date now we've moved to more stubs > +=== Rationale summary > + > +The Git Standard Library allows us to get the libification ball rolling > +with other libraries in Git. By not spending many more months attempting > +to refactor difficult circular dependencies and instead spending that > +time getting to a state where we can test out swapping a library out > +such as config or object store, we can prove the viability of Git > +libification on a much faster time scale. Additionally the code cleanups > +that have happened so far have been minor and beneficial for the > +codebase. It is probable that making large movements would negatively > +affect code clarity. > + > +== Git Standard Library boundary > + > +While I have described above some useful heuristics for identifying > +potential candidates for git-std-lib.a, a standard library should not > +have a shaky definition for what belongs in it. Maybe "we need a more precise definition" rather than the "shaky definition" bit > + - Low-level files (aka operates only on other primitive types) that are > + used everywhere within the codebase (wrapper.c, usage.c, strbuf.c) > + - Dependencies that are low-level and widely used > + (abspath.c, date.c, hex-ll.c, parse.c, utf8.c) > + - low-level git/* files with functions defined in git-compat-util.h > + (ctype.c) > + - compat/* > + > +There are other files that might fit this definition, but that does not > +mean it should belong in git-std-lib.a. Those files should start as > +their own separate library since any file added to git-std-lib.a loses > +its flexibility of being easily swappable. > + > +Wrapper.c and usage.c have dependencies on pager and trace2 that are > +possible to remove at the cost of sacrificing the ability for standard Git > +to be able to trace functions in those files and other files in git-std-lib.a. > +In order for git-std-lib.a to compile with those dependencies, stubbed out > +versions of those files are implemented and swapped in during compilation time > +(see STUB_LIB_OBJS in the Makefile). > + > +== Files inside of Git Standard Library > + > +The set of files in git-std-lib.a can be found in STD_LIB_OBJS and COMPAT_OBJS > +in the Makefile. > + > +When these files are compiled together with the files in STUB_LIB_OBJS (or > +user-provided files that provide the same functions), they form a complete > +library. > + > +== Pitfalls > + > +There are a small amount of files under compat/* that have dependencies s/amount/number/ as files are countable > +not inside of git-std-lib.a. While those functions are not called on > +Linux, other OSes might call those problematic functions. I don't see > +this as a major problem, just moreso an observation that libification in > +general may also require some minor compatibility work in the future. > + > +== Testing > + > +Unit tests should catch any breakages caused by changes to files in > +git-std-lib.a (i.e. introduction of a out of scope dependency) and new > +functions introduced to git-std-lib.a will require unit tests written s/test written/tests to be written/ > +for them. > + > +[1] https://lore.kernel.org/git/xmqqwn17sydw.fsf@gitster.g/ It is nice to see us compiling git using git-std-lib. Best Wishes Phillip > diff --git a/Makefile b/Makefile > index 4e255c81f2..d37ea9d34b 100644 > --- a/Makefile > +++ b/Makefile > @@ -669,6 +669,8 @@ FUZZ_PROGRAMS = > GIT_OBJS = > LIB_OBJS = > SCALAR_OBJS = > +STD_LIB_OBJS = > +STUB_LIB_OBJS = > OBJECTS = > OTHER_PROGRAMS = > PROGRAM_OBJS = > @@ -923,6 +925,8 @@ TEST_SHELL_PATH = $(SHELL_PATH) > > LIB_FILE = libgit.a > XDIFF_LIB = xdiff/lib.a > +STD_LIB_FILE = git-std-lib.a > +STUB_LIB_FILE = git-stub-lib.a > REFTABLE_LIB = reftable/libreftable.a > REFTABLE_TEST_LIB = reftable/libreftable_test.a > > @@ -962,7 +966,6 @@ COCCI_SOURCES = $(filter-out $(THIRD_PARTY_SOURCES),$(FOUND_C_SOURCES)) > > LIB_H = $(FOUND_H_SOURCES) > > -LIB_OBJS += abspath.o > LIB_OBJS += add-interactive.o > LIB_OBJS += add-patch.o > LIB_OBJS += advice.o > @@ -1004,8 +1007,6 @@ LIB_OBJS += convert.o > LIB_OBJS += copy.o > LIB_OBJS += credential.o > LIB_OBJS += csum-file.o > -LIB_OBJS += ctype.o > -LIB_OBJS += date.o > LIB_OBJS += decorate.o > LIB_OBJS += delta-islands.o > LIB_OBJS += diagnose.o > @@ -1046,7 +1047,6 @@ LIB_OBJS += hash-lookup.o > LIB_OBJS += hashmap.o > LIB_OBJS += help.o > LIB_OBJS += hex.o > -LIB_OBJS += hex-ll.o > LIB_OBJS += hook.o > LIB_OBJS += ident.o > LIB_OBJS += json-writer.o > @@ -1097,7 +1097,6 @@ LIB_OBJS += pack-write.o > LIB_OBJS += packfile.o > LIB_OBJS += pager.o > LIB_OBJS += parallel-checkout.o > -LIB_OBJS += parse.o > LIB_OBJS += parse-options-cb.o > LIB_OBJS += parse-options.o > LIB_OBJS += patch-delta.o > @@ -1152,7 +1151,6 @@ LIB_OBJS += sparse-index.o > LIB_OBJS += split-index.o > LIB_OBJS += stable-qsort.o > LIB_OBJS += statinfo.o > -LIB_OBJS += strbuf.o > LIB_OBJS += streaming.o > LIB_OBJS += string-list.o > LIB_OBJS += strmap.o > @@ -1189,21 +1187,32 @@ LIB_OBJS += unpack-trees.o > LIB_OBJS += upload-pack.o > LIB_OBJS += url.o > LIB_OBJS += urlmatch.o > -LIB_OBJS += usage.o > LIB_OBJS += userdiff.o > -LIB_OBJS += utf8.o > LIB_OBJS += varint.o > LIB_OBJS += version.o > LIB_OBJS += versioncmp.o > LIB_OBJS += walker.o > LIB_OBJS += wildmatch.o > LIB_OBJS += worktree.o > -LIB_OBJS += wrapper.o > LIB_OBJS += write-or-die.o > LIB_OBJS += ws.o > LIB_OBJS += wt-status.o > LIB_OBJS += xdiff-interface.o > > +STD_LIB_OBJS += abspath.o > +STD_LIB_OBJS += ctype.o > +STD_LIB_OBJS += date.o > +STD_LIB_OBJS += hex-ll.o > +STD_LIB_OBJS += parse.o > +STD_LIB_OBJS += strbuf.o > +STD_LIB_OBJS += usage.o > +STD_LIB_OBJS += utf8.o > +STD_LIB_OBJS += wrapper.o > + > +STUB_LIB_OBJS += stubs/trace2.o > +STUB_LIB_OBJS += stubs/pager.o > +STUB_LIB_OBJS += stubs/misc.o > + > BUILTIN_OBJS += builtin/add.o > BUILTIN_OBJS += builtin/am.o > BUILTIN_OBJS += builtin/annotate.o > @@ -1352,7 +1361,7 @@ UNIT_TEST_OBJS = $(patsubst %,$(UNIT_TEST_DIR)/%.o,$(UNIT_TEST_PROGRAMS)) > UNIT_TEST_OBJS += $(UNIT_TEST_DIR)/test-lib.o > > # xdiff and reftable libs may in turn depend on what is in libgit.a > -GITLIBS = common-main.o $(LIB_FILE) $(XDIFF_LIB) $(REFTABLE_LIB) $(LIB_FILE) > +GITLIBS = common-main.o $(STD_LIB_FILE) $(LIB_FILE) $(XDIFF_LIB) $(REFTABLE_LIB) $(LIB_FILE) > EXTLIBS = > > GIT_USER_AGENT = git/$(GIT_VERSION) > @@ -2693,6 +2702,8 @@ OBJECTS += $(XDIFF_OBJS) > OBJECTS += $(FUZZ_OBJS) > OBJECTS += $(REFTABLE_OBJS) $(REFTABLE_TEST_OBJS) > OBJECTS += $(UNIT_TEST_OBJS) > +OBJECTS += $(STD_LIB_OBJS) > +OBJECTS += $(STUB_LIB_OBJS) > > ifndef NO_CURL > OBJECTS += http.o http-walker.o remote-curl.o > @@ -3686,7 +3697,7 @@ clean: profile-clean coverage-clean cocciclean > $(RM) git.res > $(RM) $(OBJECTS) > $(RM) headless-git.o > - $(RM) $(LIB_FILE) $(XDIFF_LIB) $(REFTABLE_LIB) $(REFTABLE_TEST_LIB) > + $(RM) $(LIB_FILE) $(XDIFF_LIB) $(REFTABLE_LIB) $(REFTABLE_TEST_LIB) $(STD_LIB_FILE) $(STUB_LIB_FILE) > $(RM) $(ALL_PROGRAMS) $(SCRIPT_LIB) $(BUILT_INS) $(OTHER_PROGRAMS) > $(RM) $(TEST_PROGRAMS) > $(RM) $(FUZZ_PROGRAMS) > @@ -3878,3 +3889,18 @@ $(UNIT_TEST_PROGS): $(UNIT_TEST_BIN)/%$X: $(UNIT_TEST_DIR)/%.o $(UNIT_TEST_DIR)/ > build-unit-tests: $(UNIT_TEST_PROGS) > unit-tests: $(UNIT_TEST_PROGS) > $(MAKE) -C t/ unit-tests > + > +### Libified Git rules > + > +# git-std-lib.a > +# Programs other than git should compile this with > +# make NO_GETTEXT=YesPlease git-std-lib.a > +# and link against git-stub-lib.a (if the default no-op functionality is fine) > +# or a custom .a file with the same interface as git-stub-lib.a (if custom > +# functionality is needed) as well. > +$(STD_LIB_FILE): $(STD_LIB_OBJS) $(COMPAT_OBJS) > + $(QUIET_AR)$(RM) $@ && $(AR) $(ARFLAGS) $@ $^ > + > +# git-stub-lib.a > +$(STUB_LIB_FILE): $(STUB_LIB_OBJS) > + $(QUIET_AR)$(RM) $@ && $(AR) $(ARFLAGS) $@ $^ > diff --git a/stubs/misc.c b/stubs/misc.c > new file mode 100644 > index 0000000000..92da76fd46 > --- /dev/null > +++ b/stubs/misc.c > @@ -0,0 +1,33 @@ > +#include <assert.h> > +#include <stdlib.h> > + > +#ifndef NO_GETTEXT > +/* > + * NEEDSWORK: This is enough to link our unit tests against > + * git-std-lib.a built with gettext support. We don't really support > + * programs other than git using git-std-lib.a with gettext support > + * yet. To do that we need to start using dgettext() rather than > + * gettext() in our code. > + */ > +int git_gettext_enabled = 0; > +#endif > + > +int common_exit(const char *file, int line, int code); > + > +int common_exit(const char *file, int line, int code) > +{ > + exit(code); > +} > + > +#if !defined(__MINGW32__) && !defined(_MSC_VER) > +int lstat_cache_aware_rmdir(const char *path); > + > +int lstat_cache_aware_rmdir(const char *path) > +{ > + /* > + * This function should not be called by programs linked > + * against git-stub-lib.a > + */ > + assert(0); > +} > +#endif > diff --git a/stubs/pager.c b/stubs/pager.c > new file mode 100644 > index 0000000000..4f575cada7 > --- /dev/null > +++ b/stubs/pager.c > @@ -0,0 +1,6 @@ > +#include "pager.h" > + > +int pager_in_use(void) > +{ > + return 0; > +} > diff --git a/stubs/trace2.c b/stubs/trace2.c > new file mode 100644 > index 0000000000..7d89482228 > --- /dev/null > +++ b/stubs/trace2.c > @@ -0,0 +1,27 @@ > +#include "git-compat-util.h" > +#include "trace2.h" > + > +struct child_process { int stub; }; > +struct repository { int stub; }; > +struct json_writer { int stub; }; > + > +void trace2_region_enter_fl(const char *file, int line, const char *category, > + const char *label, const struct repository *repo, ...) { } > +void trace2_region_leave_fl(const char *file, int line, const char *category, > + const char *label, const struct repository *repo, ...) { } > +void trace2_data_string_fl(const char *file, int line, const char *category, > + const struct repository *repo, const char *key, > + const char *value) { } > +void trace2_cmd_ancestry_fl(const char *file, int line, const char **parent_names) { } > +void trace2_cmd_error_va_fl(const char *file, int line, const char *fmt, > + va_list ap) { } > +void trace2_cmd_name_fl(const char *file, int line, const char *name) { } > +void trace2_thread_start_fl(const char *file, int line, > + const char *thread_base_name) { } > +void trace2_thread_exit_fl(const char *file, int line) { } > +void trace2_data_intmax_fl(const char *file, int line, const char *category, > + const struct repository *repo, const char *key, > + intmax_t value) { } > +int trace2_is_enabled(void) { return 0; } > +void trace2_counter_add(enum trace2_counter_id cid, uint64_t value) { } > +void trace2_collect_process_info(enum trace2_process_info_reason reason) { }
Phillip Wood <phillip.wood123@gmail.com> writes: > Hi Calvin > On 22/02/2024 17:50, Calvin Wan wrote: Thanks for reviewing -- I held mine back as I expected this will see another reroll soonish, but you already have raised many points I also had trouble with, so I do not have to ;-) Below, I'll liberally omit everything you wrote that I agree with. >> +libgit.a, which is then linked against by common-main.o with other >> +external dependencies and turned into the Git executable. > > I found this description a bit confusing. As I understand it to build > git we link git.o against common-main.o, libgit.a, xdiff/lib.a, > reftable/libreftable.a and libpcre etc. In addition, there is no single "the Git executable", simply because not everything is builtin command. The purpose of using libgit.a is because we are too lazy to list and maintain all the internal dependencies to link final executables like 'git' (which has all the built-in command implementations) and 'git-remote-curl' (which is a standalone program). Instead of feeding exact list of object files to "$(CC) -o git" command line, we throw everything into libgit.a and let the linker pick what is needed. To link "git", we may include all builtin/*.o, git.o, common-main.o, libgit.a and the external [*] library dependencies they have. To link "git-daemon", we may not link builtin/*.o and git.o and link daemon.o instead. Side note: here I am counting xdiff/lib.a as an external library as it is mostly a borrowed code. In other words, libgit.a is not a true library in the sense that it was designed to be _used_ as a library. It was merely a detail of how we implemented lazy dependency management in our build process, which happend with 0a02ce72 (Clean up the Makefile a bit., 2005-04-18) whose commit log message uses air-quotes around the word "library". >> +From the above dependency graph, we can see that git-std-lib.a could be >> +many smaller libraries rather than a singular library. So why choose a >> +singular library when multiple libraries can be individually easier to >> +swap and are more modular? A singular library requires less work to >> +separate out circular dependencies within itself so it becomes a >> +tradeoff question between work and reward. While there may be a point in >> +the future where a file like usage.c would want its own library so that >> +someone can have custom die() or error(), the work required to refactor >> +out the circular dependencies in some files would be enormous due to >> +their ubiquity so therefore I believe it is not worth the tradeoff > > I'm not sure if we want to use the first person in our technical > documentation, unlike the cover letter to a patch series it is not > immediately obvious to the reader who "I" is. This applies the > passages in the first person below as well. I found it highly annoying while reading it, too. If the document (not the commit that introduced the document) were signed and written to state the position of one author, as opposed to spell out the position the project will collectively take, it would have been OK, but this document is meant to set a course of the project (and discussion on it is the process to decide which course to take), the first person singular "I" did not sit well for me. Another thing to consider that I do not think you covered is the name of the resulting .a archive. By starting it with "lib", a customer can find your libstdgit.a with -lstdgit on the command line, once libstdgit.a is installed at an appropriate location (or the build-time library path is configured to point the location you have libstdgit.a at). libgit.a that wasn't really designed to be used as such a library did not have to follow the naming convention, but if the thing being proposed is meant to be eventually used as a library by external entities, git-std-lib.a is a rather poor name for it. PS. I seem to have been hit by a power outage and am on UPS, so I'll probably be offline until the power comes back.
Junio C Hamano <gitster@pobox.com> writes: > Phillip Wood <phillip.wood123@gmail.com> writes: > >> On 22/02/2024 17:50, Calvin Wan wrote: > >>> +libgit.a, which is then linked against by common-main.o with other >>> +external dependencies and turned into the Git executable. >> >> I found this description a bit confusing. As I understand it to build >> git we link git.o against common-main.o, libgit.a, xdiff/lib.a, >> reftable/libreftable.a and libpcre etc. > > In addition, there is no single "the Git executable", simply because > not everything is builtin command. The purpose of using libgit.a is > because we are too lazy to list and maintain all the internal > dependencies to link final executables like 'git' (which has all the > built-in command implementations) and 'git-remote-curl' (which is a > standalone program). Instead of feeding exact list of object files > to "$(CC) -o git" command line, we throw everything into libgit.a > and let the linker pick what is needed. To link "git", we may > include all builtin/*.o, git.o, common-main.o, libgit.a and the > external [*] library dependencies they have. To link "git-daemon", > we may not link builtin/*.o and git.o and link daemon.o instead. > > Side note: here I am counting xdiff/lib.a as an external > library as it is mostly a borrowed code. > > In other words, libgit.a is not a true library in the sense that it > was designed to be _used_ as a library. It was merely a detail of > how we implemented lazy dependency management in our build process, > which happend with 0a02ce72 (Clean up the Makefile a bit., > 2005-04-18) whose commit log message uses air-quotes around the word > "library". Somehow I did not realize that this was going on. Thank you for pointing this out! It does make me wonder if we should stop being lazy and do the work that the linker has been doing for us "for free" ourselves. IOW, stop linking against a monolithic libgit.a. That way we would replace implicit dependencies with explicit ones, which might help us understand which things need what. But maybe doing that is super painful, so, maybe it's not worth it. IDK.
Linus Arver <linusa@google.com> writes: > It does make me wonder if we should stop being lazy and do the > work that the linker has been doing for us "for free" > ourselves. IOW, stop linking against a monolithic libgit.a. > ... which might help us understand which things need what. Sorry, but I fail see a point in such an exercise. If a tool is available to help us and if there is no downside of using the tool, we should keep using it. If you are proposing to move away from the current build practice because you have a concrete downside of the approach and avoid that, then it might be a good proposal, though. And "we do not learn otherwise" is not a downside of the approach; "we do not learn" comes from your not learning, the tools do not force you to be ignorant. We do not propose to use more __asm__ in our C sources only because compilers were doing that for us "for free" and because the compilers were somehow robbing us the opportunity to learn micro-optimization techniques, do we? A small downside I can immediately think of is possible in a situation where we have been throwing an object file into libgit.a archive that is no longer used by any final executable. In such a scenario, if you change the source file that is compiled into such an unused object file, your next "make" will update libgit.a to replace the unused object file with its new version with your updates, and that would cause the final build product to be linked again with objects needed from libgit.a, but there shouldn't be any change because we are talking about an object that is *not* used by them but still is in libgit.a due to be listed on LIB_OBJS variable. But that is a purely theoretical downside. It may be the case that we haven't done our spring cleaning recently and we haven't noticed that a source file or two are now unused but are still listed on LIB_OBJS to be included in the libgit.a archive. But even if that were the case, it is implausible that you are touching such an unused source file in the first place.
Junio C Hamano <gitster@pobox.com> writes: > Linus Arver <linusa@google.com> writes: > >> It does make me wonder if we should stop being lazy and do the >> work that the linker has been doing for us "for free" >> ourselves. IOW, stop linking against a monolithic libgit.a. >> ... which might help us understand which things need what. > > [...] If a tool is > available to help us and if there is no downside of using the tool, > we should keep using it. Of course, if there is no downside, we should use the tool as is. > If you are proposing to move away from the > current build practice because you have a concrete downside of the > approach and avoid that, then it might be a good proposal, though. Right. I was just wondering if the "explicit dependencies declared in the Makefile" would provide some value WRT libification. Currently IDK the answer to that. > And "we do not learn otherwise" is not a downside of the approach; > "we do not learn" comes from your not learning, the tools do not > force you to be ignorant. We do not propose to use more __asm__ in > our C sources only because compilers were doing that for us "for > free" and because the compilers were somehow robbing us the > opportunity to learn micro-optimization techniques, do we? True. > A small downside I can immediately think of is possible in a > situation where we have been throwing an object file into libgit.a > archive that is no longer used by any final executable. In such a > scenario, if you change the source file that is compiled into such > an unused object file, your next "make" will update libgit.a to > replace the unused object file with its new version with your > updates, and that would cause the final build product to be linked > again with objects needed from libgit.a, but there shouldn't be any > change because we are talking about an object that is *not* used by > them but still is in libgit.a due to be listed on LIB_OBJS variable. IIUC, this (theoretical) downside will result in Make thinking that it needs to rebuild libgit.a when it actually doesn't need to (because the updated change is for an unused object). So it could slow down the build unnecessarily. Makes sense. > But that is a purely theoretical downside. It may be the case that > we haven't done our spring cleaning recently and we haven't noticed > that a source file or two are now unused but are still listed on > LIB_OBJS to be included in the libgit.a archive. But even if that > were the case, it is implausible that you are touching such an > unused source file in the first place. As you noted, libgit.a is not a true library; it's just a big archive of everything and we let the linker figure out what the executables need out of it. But I was under the impression that with Git libification, we would want to create a real library in the fullest sense of the word --- such that our executables also need this library ("-lgit") to be built in the exact same way that external programs would need this same (single) Git library. For example, I believe this is how curl is built (it first builds libcurl, and then links against it as an internal user for generating the "curl" executable). Going back to libgit.a, I was just wondering if the exercise of breaking it up into smaller pieces would have been helpful in figuring out this "-lgit" library (or what a smaller version of it would look like). I sense that I may be missing large pieces of context around the git-std-lib discussions though, so I apologize if my points above are not new and moot. Thanks.
diff --git a/Documentation/Makefile b/Documentation/Makefile index 3f2383a12c..f1dc673838 100644 --- a/Documentation/Makefile +++ b/Documentation/Makefile @@ -110,6 +110,7 @@ TECH_DOCS += SubmittingPatches TECH_DOCS += ToolsForGit TECH_DOCS += technical/bitmap-format TECH_DOCS += technical/bundle-uri +TECH_DOCS += technical/git-std-lib TECH_DOCS += technical/hash-function-transition TECH_DOCS += technical/long-running-process-protocol TECH_DOCS += technical/multi-pack-index diff --git a/Documentation/technical/git-std-lib.txt b/Documentation/technical/git-std-lib.txt new file mode 100644 index 0000000000..3d9aa121ac --- /dev/null +++ b/Documentation/technical/git-std-lib.txt @@ -0,0 +1,170 @@ += Git Standard Library + +The Git Standard Library intends to serve as the foundational library +and root dependency that other libraries in Git will be built off of. +That is to say, suppose we have libraries X and Y; a user that wants to +use X and Y would need to include X, Y, and this Git Standard Library. +This does not mean that the Git Standard Library will be the only +possible root dependency in the future, but rather the most significant +and widely used one. Git itself is also built off of the Git Standard +Library. + +== Dependency graph in libified Git + +Before the introduction of the Git Standard Library, all objects defined +in the Git library are compiled and archived into a singular file, +libgit.a, which is then linked against by common-main.o with other +external dependencies and turned into the Git executable. In other +words, the Git executable has dependencies on libgit.a and a couple of +external libraries. The libfication of Git slightly alters this build +flow by separating out libgit.a into libgit.a and git-std-lib.a. + +With our current method of building Git, we can imagine the dependency +graph as such: + + Git + /\ + / \ + / \ + libgit.a ext deps + +We want to separate out potential libraries from libgit.a and have +libgit.a depend on them, which would possibly look like: + + Git + /\ + / \ + / \ + libgit.a ext deps + /\ + / \ + / \ +object-store.a (other lib) + | / + | / + | / + | / + | / + | / + | / +git-std-lib.a + +Instead of containing all objects in Git, libgit.a would contain objects +that are not built by libraries it links against. Consequently, if +someone wanted a custom build of Git with a custom implementation of the +object store, they would only have to swap out object-store.a rather +than do a hard fork of Git. + +== Rationale behind Git Standard Library + +The rationale behind the selected object files in the Git Standard +Library is the result of two observations within the Git +codebase: + 1. every file includes git-compat-util.h which defines functions + in a couple of different files + 2. wrapper.c + usage.c have difficult-to-separate circular + dependencies with each other and other files. + +=== Ubiquity of git-compat-util.h and circular dependencies + +Every file in the Git codebase includes git-compat-util.h. It serves as +"a compatibility aid that isolates the knowledge of platform specific +inclusion order and what feature macros to define before including which +system header" (Junio[1]). Since every file includes git-compat-util.h, +and git-compat-util.h includes wrapper.h and usage.h, it would make +sense for wrapper.c and usage.c to be a part of the root library. They +have difficult to separate circular dependencies with each other so it +would impractical for them to be independent libraries. Wrapper.c has +dependencies on parse.c, abspath.c, strbuf.c, which in turn also have +dependencies on usage.c and wrapper.c - more circular dependencies. + +=== Tradeoff between swappability and refactoring + +From the above dependency graph, we can see that git-std-lib.a could be +many smaller libraries rather than a singular library. So why choose a +singular library when multiple libraries can be individually easier to +swap and are more modular? A singular library requires less work to +separate out circular dependencies within itself so it becomes a +tradeoff question between work and reward. While there may be a point in +the future where a file like usage.c would want its own library so that +someone can have custom die() or error(), the work required to refactor +out the circular dependencies in some files would be enormous due to +their ubiquity so therefore I believe it is not worth the tradeoff +currently. Additionally, we can in the future choose to do this refactor +and change the API for the library if there becomes enough of a reason +to do so (remember we are avoiding promising stability of the interfaces +of those libraries). + +=== Reuse of compatibility functions in git-compat-util.h + +Most functions defined in git-compat-util.h are implemented in compat/ +and have dependencies limited to strbuf.h and wrapper.h so they can be +easily included in git-std-lib.a, which as a root dependency means that +higher level libraries do not have to worry about compatibility files in +compat/. The rest of the functions defined in git-compat-util.h are +implemented in top level files and are hidden behind +an #ifdef if their implementation is not in git-std-lib.a. + +=== Rationale summary + +The Git Standard Library allows us to get the libification ball rolling +with other libraries in Git. By not spending many more months attempting +to refactor difficult circular dependencies and instead spending that +time getting to a state where we can test out swapping a library out +such as config or object store, we can prove the viability of Git +libification on a much faster time scale. Additionally the code cleanups +that have happened so far have been minor and beneficial for the +codebase. It is probable that making large movements would negatively +affect code clarity. + +== Git Standard Library boundary + +While I have described above some useful heuristics for identifying +potential candidates for git-std-lib.a, a standard library should not +have a shaky definition for what belongs in it. + + - Low-level files (aka operates only on other primitive types) that are + used everywhere within the codebase (wrapper.c, usage.c, strbuf.c) + - Dependencies that are low-level and widely used + (abspath.c, date.c, hex-ll.c, parse.c, utf8.c) + - low-level git/* files with functions defined in git-compat-util.h + (ctype.c) + - compat/* + +There are other files that might fit this definition, but that does not +mean it should belong in git-std-lib.a. Those files should start as +their own separate library since any file added to git-std-lib.a loses +its flexibility of being easily swappable. + +Wrapper.c and usage.c have dependencies on pager and trace2 that are +possible to remove at the cost of sacrificing the ability for standard Git +to be able to trace functions in those files and other files in git-std-lib.a. +In order for git-std-lib.a to compile with those dependencies, stubbed out +versions of those files are implemented and swapped in during compilation time +(see STUB_LIB_OBJS in the Makefile). + +== Files inside of Git Standard Library + +The set of files in git-std-lib.a can be found in STD_LIB_OBJS and COMPAT_OBJS +in the Makefile. + +When these files are compiled together with the files in STUB_LIB_OBJS (or +user-provided files that provide the same functions), they form a complete +library. + +== Pitfalls + +There are a small amount of files under compat/* that have dependencies +not inside of git-std-lib.a. While those functions are not called on +Linux, other OSes might call those problematic functions. I don't see +this as a major problem, just moreso an observation that libification in +general may also require some minor compatibility work in the future. + +== Testing + +Unit tests should catch any breakages caused by changes to files in +git-std-lib.a (i.e. introduction of a out of scope dependency) and new +functions introduced to git-std-lib.a will require unit tests written +for them. + +[1] https://lore.kernel.org/git/xmqqwn17sydw.fsf@gitster.g/ diff --git a/Makefile b/Makefile index 4e255c81f2..d37ea9d34b 100644 --- a/Makefile +++ b/Makefile @@ -669,6 +669,8 @@ FUZZ_PROGRAMS = GIT_OBJS = LIB_OBJS = SCALAR_OBJS = +STD_LIB_OBJS = +STUB_LIB_OBJS = OBJECTS = OTHER_PROGRAMS = PROGRAM_OBJS = @@ -923,6 +925,8 @@ TEST_SHELL_PATH = $(SHELL_PATH) LIB_FILE = libgit.a XDIFF_LIB = xdiff/lib.a +STD_LIB_FILE = git-std-lib.a +STUB_LIB_FILE = git-stub-lib.a REFTABLE_LIB = reftable/libreftable.a REFTABLE_TEST_LIB = reftable/libreftable_test.a @@ -962,7 +966,6 @@ COCCI_SOURCES = $(filter-out $(THIRD_PARTY_SOURCES),$(FOUND_C_SOURCES)) LIB_H = $(FOUND_H_SOURCES) -LIB_OBJS += abspath.o LIB_OBJS += add-interactive.o LIB_OBJS += add-patch.o LIB_OBJS += advice.o @@ -1004,8 +1007,6 @@ LIB_OBJS += convert.o LIB_OBJS += copy.o LIB_OBJS += credential.o LIB_OBJS += csum-file.o -LIB_OBJS += ctype.o -LIB_OBJS += date.o LIB_OBJS += decorate.o LIB_OBJS += delta-islands.o LIB_OBJS += diagnose.o @@ -1046,7 +1047,6 @@ LIB_OBJS += hash-lookup.o LIB_OBJS += hashmap.o LIB_OBJS += help.o LIB_OBJS += hex.o -LIB_OBJS += hex-ll.o LIB_OBJS += hook.o LIB_OBJS += ident.o LIB_OBJS += json-writer.o @@ -1097,7 +1097,6 @@ LIB_OBJS += pack-write.o LIB_OBJS += packfile.o LIB_OBJS += pager.o LIB_OBJS += parallel-checkout.o -LIB_OBJS += parse.o LIB_OBJS += parse-options-cb.o LIB_OBJS += parse-options.o LIB_OBJS += patch-delta.o @@ -1152,7 +1151,6 @@ LIB_OBJS += sparse-index.o LIB_OBJS += split-index.o LIB_OBJS += stable-qsort.o LIB_OBJS += statinfo.o -LIB_OBJS += strbuf.o LIB_OBJS += streaming.o LIB_OBJS += string-list.o LIB_OBJS += strmap.o @@ -1189,21 +1187,32 @@ LIB_OBJS += unpack-trees.o LIB_OBJS += upload-pack.o LIB_OBJS += url.o LIB_OBJS += urlmatch.o -LIB_OBJS += usage.o LIB_OBJS += userdiff.o -LIB_OBJS += utf8.o LIB_OBJS += varint.o LIB_OBJS += version.o LIB_OBJS += versioncmp.o LIB_OBJS += walker.o LIB_OBJS += wildmatch.o LIB_OBJS += worktree.o -LIB_OBJS += wrapper.o LIB_OBJS += write-or-die.o LIB_OBJS += ws.o LIB_OBJS += wt-status.o LIB_OBJS += xdiff-interface.o +STD_LIB_OBJS += abspath.o +STD_LIB_OBJS += ctype.o +STD_LIB_OBJS += date.o +STD_LIB_OBJS += hex-ll.o +STD_LIB_OBJS += parse.o +STD_LIB_OBJS += strbuf.o +STD_LIB_OBJS += usage.o +STD_LIB_OBJS += utf8.o +STD_LIB_OBJS += wrapper.o + +STUB_LIB_OBJS += stubs/trace2.o +STUB_LIB_OBJS += stubs/pager.o +STUB_LIB_OBJS += stubs/misc.o + BUILTIN_OBJS += builtin/add.o BUILTIN_OBJS += builtin/am.o BUILTIN_OBJS += builtin/annotate.o @@ -1352,7 +1361,7 @@ UNIT_TEST_OBJS = $(patsubst %,$(UNIT_TEST_DIR)/%.o,$(UNIT_TEST_PROGRAMS)) UNIT_TEST_OBJS += $(UNIT_TEST_DIR)/test-lib.o # xdiff and reftable libs may in turn depend on what is in libgit.a -GITLIBS = common-main.o $(LIB_FILE) $(XDIFF_LIB) $(REFTABLE_LIB) $(LIB_FILE) +GITLIBS = common-main.o $(STD_LIB_FILE) $(LIB_FILE) $(XDIFF_LIB) $(REFTABLE_LIB) $(LIB_FILE) EXTLIBS = GIT_USER_AGENT = git/$(GIT_VERSION) @@ -2693,6 +2702,8 @@ OBJECTS += $(XDIFF_OBJS) OBJECTS += $(FUZZ_OBJS) OBJECTS += $(REFTABLE_OBJS) $(REFTABLE_TEST_OBJS) OBJECTS += $(UNIT_TEST_OBJS) +OBJECTS += $(STD_LIB_OBJS) +OBJECTS += $(STUB_LIB_OBJS) ifndef NO_CURL OBJECTS += http.o http-walker.o remote-curl.o @@ -3686,7 +3697,7 @@ clean: profile-clean coverage-clean cocciclean $(RM) git.res $(RM) $(OBJECTS) $(RM) headless-git.o - $(RM) $(LIB_FILE) $(XDIFF_LIB) $(REFTABLE_LIB) $(REFTABLE_TEST_LIB) + $(RM) $(LIB_FILE) $(XDIFF_LIB) $(REFTABLE_LIB) $(REFTABLE_TEST_LIB) $(STD_LIB_FILE) $(STUB_LIB_FILE) $(RM) $(ALL_PROGRAMS) $(SCRIPT_LIB) $(BUILT_INS) $(OTHER_PROGRAMS) $(RM) $(TEST_PROGRAMS) $(RM) $(FUZZ_PROGRAMS) @@ -3878,3 +3889,18 @@ $(UNIT_TEST_PROGS): $(UNIT_TEST_BIN)/%$X: $(UNIT_TEST_DIR)/%.o $(UNIT_TEST_DIR)/ build-unit-tests: $(UNIT_TEST_PROGS) unit-tests: $(UNIT_TEST_PROGS) $(MAKE) -C t/ unit-tests + +### Libified Git rules + +# git-std-lib.a +# Programs other than git should compile this with +# make NO_GETTEXT=YesPlease git-std-lib.a +# and link against git-stub-lib.a (if the default no-op functionality is fine) +# or a custom .a file with the same interface as git-stub-lib.a (if custom +# functionality is needed) as well. +$(STD_LIB_FILE): $(STD_LIB_OBJS) $(COMPAT_OBJS) + $(QUIET_AR)$(RM) $@ && $(AR) $(ARFLAGS) $@ $^ + +# git-stub-lib.a +$(STUB_LIB_FILE): $(STUB_LIB_OBJS) + $(QUIET_AR)$(RM) $@ && $(AR) $(ARFLAGS) $@ $^ diff --git a/stubs/misc.c b/stubs/misc.c new file mode 100644 index 0000000000..92da76fd46 --- /dev/null +++ b/stubs/misc.c @@ -0,0 +1,33 @@ +#include <assert.h> +#include <stdlib.h> + +#ifndef NO_GETTEXT +/* + * NEEDSWORK: This is enough to link our unit tests against + * git-std-lib.a built with gettext support. We don't really support + * programs other than git using git-std-lib.a with gettext support + * yet. To do that we need to start using dgettext() rather than + * gettext() in our code. + */ +int git_gettext_enabled = 0; +#endif + +int common_exit(const char *file, int line, int code); + +int common_exit(const char *file, int line, int code) +{ + exit(code); +} + +#if !defined(__MINGW32__) && !defined(_MSC_VER) +int lstat_cache_aware_rmdir(const char *path); + +int lstat_cache_aware_rmdir(const char *path) +{ + /* + * This function should not be called by programs linked + * against git-stub-lib.a + */ + assert(0); +} +#endif diff --git a/stubs/pager.c b/stubs/pager.c new file mode 100644 index 0000000000..4f575cada7 --- /dev/null +++ b/stubs/pager.c @@ -0,0 +1,6 @@ +#include "pager.h" + +int pager_in_use(void) +{ + return 0; +} diff --git a/stubs/trace2.c b/stubs/trace2.c new file mode 100644 index 0000000000..7d89482228 --- /dev/null +++ b/stubs/trace2.c @@ -0,0 +1,27 @@ +#include "git-compat-util.h" +#include "trace2.h" + +struct child_process { int stub; }; +struct repository { int stub; }; +struct json_writer { int stub; }; + +void trace2_region_enter_fl(const char *file, int line, const char *category, + const char *label, const struct repository *repo, ...) { } +void trace2_region_leave_fl(const char *file, int line, const char *category, + const char *label, const struct repository *repo, ...) { } +void trace2_data_string_fl(const char *file, int line, const char *category, + const struct repository *repo, const char *key, + const char *value) { } +void trace2_cmd_ancestry_fl(const char *file, int line, const char **parent_names) { } +void trace2_cmd_error_va_fl(const char *file, int line, const char *fmt, + va_list ap) { } +void trace2_cmd_name_fl(const char *file, int line, const char *name) { } +void trace2_thread_start_fl(const char *file, int line, + const char *thread_base_name) { } +void trace2_thread_exit_fl(const char *file, int line) { } +void trace2_data_intmax_fl(const char *file, int line, const char *category, + const struct repository *repo, const char *key, + intmax_t value) { } +int trace2_is_enabled(void) { return 0; } +void trace2_counter_add(enum trace2_counter_id cid, uint64_t value) { } +void trace2_collect_process_info(enum trace2_process_info_reason reason) { }