| Message ID | 20240426152449.228860-4-knayak@gitlab.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | add symref-* commands to 'git-update-ref --stdin' | expand |
Karthik Nayak <karthik.188@gmail.com> writes: > From: Karthik Nayak <karthik.188@gmail.com> > > In the previous commits, we added the required base for adding symref > commands to the '--stdin' mode provided by 'git-update-ref(1)'. Using > them, add a new 'symref-verify' command to verify symrefs. > > The 'symref-verify' command allows users to verify if a provided <ref> > contains the provided <old-target> without changing the <ref>. If > <old-target> is not provided, the command will verify that the <ref> > doesn't exist. Since we're checking for symbolic refs, this command will > only work with the 'no-deref' mode. This is because any dereferenced > symbolic ref will point to an object and not a ref and the regular > 'verify' command can be used in such situations. All makes sense, but a naïve reader may find it helpful if you explained why having "verify" command is a good idea in the first place ("I can just do 'git symoblic-ref' to read the current value, and see if it is what I expect"). Presumably the value of "verify" is that you can have it in a transaction and fail other operations in the same transaction if the symref moved from what you expected it to point at? > Add and use `ref_update_is_null_new_value`, a helper function which is > used to check if there is a new_value in a reference update. The new > value could either be a symref target `new_target` or a OID `new_oid`. > We also add tests to test the command in both the regular stdin mode and > also with the '-z' flag. This looks out of place, primarily because the helper function is *NOT* used in this step. Without any actual user, and with the name that says only what it checks without hinting why a caller may want to check the condition it checks, it is hard to guess if it is a good idea to have such a helper. "If a ref_update object specifies no new-oid and no new-target, it is not about updating but just validating" is how the callers are expected to use it, then instead of is_null_new_value that says what it checks, something like is_verify_only that says what the caller may want to use it for would be a more friendly name for readers and future developers. > @@ -297,11 +320,47 @@ static void parse_cmd_verify(struct ref_transaction *transaction, > die("verify %s: extra input: %s", refname, next); > > if (ref_transaction_verify(transaction, refname, &old_oid, > - update_flags, &err)) > + NULL, update_flags, &err)) > > update_flags = default_flags; > free(refname); > strbuf_release(&err); > } The only damage by this patch to parse_cmd_verify() is that ref_transaction_verify() gained another parameter NULL, but with the default "--diff-algorithm=myers" algorithm, it is very hard to see. The "--patience" algorithm does a much beter job on this hunk. And the following function is entirely new. > +static void parse_cmd_symref_verify(struct ref_transaction *transaction, > + const char *next, const char *end) > +{ > + struct strbuf err = STRBUF_INIT; > + struct object_id old_oid; > + char *refname, *old_target; > + > + if (!(update_flags & REF_NO_DEREF)) > + die("symref-verify: cannot operate with deref mode"); > + > + refname = parse_refname(&next); > + if (!refname) > + die("symref-verify: missing <ref>"); > + > + /* > + * old_ref is optional, but we want to differentiate between > + * a NULL and zero value. > + */ > + old_target = parse_next_refname(&next); > + if (!old_target) > + old_oid = *null_oid(); In many existing code paths, we do not do structure assignment like this. Instead we do oidcpy(&old_oid, null_oid()); We can see an existing example in a common context in a hunk for refs.c in this patch. > + if (*next != line_termination) > + die("symref-verify %s: extra input: %s", refname, next); > + > + if (ref_transaction_verify(transaction, refname, > + old_target ? NULL : &old_oid, > + old_target, update_flags, &err)) > + die("%s", err.buf); Are static analyzers smart enough to notice that we will not be using old_oid uninitialized here? Just wondering. Anyway. This ensures ref_transaction_verify() gets either old_target or old_oid, but never both at the same time. The caller to ref_transaction_verify() in the previous function passed NULL for old_target but it always had a non-NULL old_oid so that is perfectly fine. > + update_flags = default_flags; > + free(refname); > + free(old_target); > + strbuf_release(&err); > +} > diff --git a/refs.c b/refs.c > index 060a31616d..0e1013b5ab 100644 > --- a/refs.c > +++ b/refs.c > @@ -1217,6 +1217,8 @@ void ref_transaction_free(struct ref_transaction *transaction) > > for (i = 0; i < transaction->nr; i++) { > free(transaction->updates[i]->msg); > + free((void *)transaction->updates[i]->old_target); > + free((void *)transaction->updates[i]->new_target); > free(transaction->updates[i]); > } > free(transaction->updates); > @@ -1247,9 +1249,13 @@ struct ref_update *ref_transaction_add_update( > > update->flags = flags; > > - if (flags & REF_HAVE_NEW) > + if (new_target) > + update->new_target = xstrdup(new_target); > + if (old_target) > + update->old_target = xstrdup(old_target); Presumably "update" structure, when freshly initialized, has NULL in both of these _target members? Otherwise ref_transaction_free() would get in trouble, so double checking. > + if (new_oid && flags & REF_HAVE_NEW) > oidcpy(&update->new_oid, new_oid); > - if (flags & REF_HAVE_OLD) > + if (old_oid && flags & REF_HAVE_OLD) > oidcpy(&update->old_oid, old_oid); Since we can ask to work on a symbolic ref, new_oid / old_oid can be NULL when REF_HAVE_NEW / REF_HAVE_OLD bit is on for _target members. Makes me wonder if the code becomes easier to follow if the flag bits are split into four (_NEW -> _NEW_OID + _NEW_TARGET), but let's not worry about that for now. > @@ -1286,6 +1292,7 @@ int ref_transaction_update(struct ref_transaction *transaction, > flags &= REF_TRANSACTION_UPDATE_ALLOWED_FLAGS; > > flags |= (new_oid ? REF_HAVE_NEW : 0) | (old_oid ? REF_HAVE_OLD : 0); > + flags |= (new_target ? REF_HAVE_NEW : 0) | (old_target ? REF_HAVE_OLD : 0); > @@ -1325,14 +1332,17 @@ int ref_transaction_delete(struct ref_transaction *transaction, > int ref_transaction_verify(struct ref_transaction *transaction, > const char *refname, > const struct object_id *old_oid, > + const char *old_target, > unsigned int flags, > struct strbuf *err) > { > - if (!old_oid) > - BUG("verify called with old_oid set to NULL"); > + if (!old_target && !old_oid) > + BUG("verify called with old_oid and old_target set to NULL"); Is it normal if you get _both_ set, or is it equally a BUG()? The parse_*_verify() codepaths we saw earlier both made sure only one of the two is non-NULL, and it is unclear what should happen if both are non-NULL. > + if (old_target && !(flags & REF_NO_DEREF)) > + BUG("verify cannot operate on symrefs with deref mode"); > return ref_transaction_update(transaction, refname, > NULL, old_oid, > - NULL, NULL, > + NULL, old_target, > flags, NULL, err); > } So this queues an ref_update object whose .new_oid and .new_target are NULL, and .old_oid and .old_target are what the caller gave us to check. The NULLs in .new* members hopefully do not mean "delete this thing" ;-) > @@ -2349,6 +2359,12 @@ static int run_transaction_hook(struct ref_transaction *transaction, > for (i = 0; i < transaction->nr; i++) { > struct ref_update *update = transaction->updates[i]; > > + /* > + * Skip reference transaction for symbolic refs. > + */ > + if (update->new_target || update->old_target) > + continue; Is that a final design, or will the hooks have a chance to interfere? > diff --git a/refs/files-backend.c b/refs/files-backend.c > index 2420dac2aa..53197fa3af 100644 > --- a/refs/files-backend.c > +++ b/refs/files-backend.c > @@ -2425,6 +2425,37 @@ static const char *original_update_refname(struct ref_update *update) > return update->refname; > } > > +/* > + * Check whether the REF_HAVE_OLD and old_target values stored in > + * update are consistent with ref, which is the symbolic reference's > + * current value. If everything is OK, return 0; otherwise, write an > + * error message to err and return -1. > + */ > +static int check_old_target(struct ref_update *update, char *ref, > + struct strbuf *err) > +{ > + if (!(update->flags & REF_HAVE_OLD) || > + !strcmp(update->old_target, ref)) > + return 0; Earlier on the assignment side for "update" structure we saw above, the guard was (old_target && flags & REF_HAVE_OLD), but here we assume old_target is valid, which feels a bit asymmetric. Yes, I can see that the caller does not call us when !old_target, but still... Perhaps if ((update->flags & REF_HAVE_OLD) && !update->old_target) BUG(...); or something? Or alternatively, perhaps !!update->old_target should be the only thing we should check and ignore REF_HAVE_OLD bit? I am not sure, but it smells like that the non-NULL-ness of old_target is the only thing that matters (if it is not NULL, very early in the control flow somebody would have set REF_HAVE_OLD bit to flags, no?). It brings me back to my earlier question. Does REF_HAVE_OLD bit serve a useful purpose in this code? > + if (!strcmp(update->old_target, "")) > + strbuf_addf(err, "cannot lock ref '%s': " > + "reference already exists", > + original_update_refname(update)); > + else if (!strcmp(ref, "")) > + strbuf_addf(err, "cannot lock ref '%s': " > + "reference is missing but expected %s", > + original_update_refname(update), > + update->old_target); So... for old_target and ref, an empty string is a special value? How? Shouldn't that be documented in the comment before the function? > + else > + strbuf_addf(err, "cannot lock ref '%s': " > + "is at %s but expected %s", > + original_update_refname(update), > + ref, update->old_target); > + > + return -1; > +} > + > /* > * Check whether the REF_HAVE_OLD and old_oid values stored in update > * are consistent with oid, which is the reference's current value. If > @@ -2528,6 +2559,18 @@ static int lock_ref_for_update(struct files_ref_store *refs, > ret = TRANSACTION_GENERIC_ERROR; > goto out; > } > + } > + > + /* > + * For symref verification, we need to check the reference value > + * rather than the oid. If we're dealing with regular refs or we're > + * verifying a dereferenced symref, we then check the oid. > + */ > + if (update->old_target) { > + if (check_old_target(update, referent.buf, err)) { > + ret = TRANSACTION_GENERIC_ERROR; > + goto out; > + } We come here only when update->type has REF_ISSYMREF bit on (we learned that value by calling lock_raw_ref()), and know referent.buf has the current "target" value. That is consumed as "ref" parameter to check_old_target() we just saw. OK. > diff --git a/refs/reftable-backend.c b/refs/reftable-backend.c > index 6104471199..a2474245aa 100644 > --- a/refs/reftable-backend.c > +++ b/refs/reftable-backend.c > @@ -938,7 +938,26 @@ static int reftable_be_transaction_prepare(struct ref_store *ref_store, > * individual refs. But the error messages match what the files > * backend returns, which keeps our tests happy. > */ > - if (u->flags & REF_HAVE_OLD && !oideq(¤t_oid, &u->old_oid)) { > + if ((u->flags & REF_HAVE_OLD) && u->old_target) { > + if (strcmp(referent.buf, u->old_target)) { > + if (!strcmp(u->old_target, "")) > + strbuf_addf(err, "verifying symref target: '%s': " > + "provided target is empty", > + original_update_refname(u)); > + else if (!strcmp(referent.buf, "")) > + strbuf_addf(err, "verifying symref target: '%s': " > + "reference is missing but expected %s", > + original_update_refname(u), > + u->old_target); > + else > + strbuf_addf(err, "verifying symref target: '%s': " > + "is at %s but expected %s", > + original_update_refname(u), > + referent.buf, u->old_target); > + ret = -1; > + goto done; > + } Again, the puzzling "empty string"s are handled here.
Junio C Hamano <gitster@pobox.com> writes: > Karthik Nayak <karthik.188@gmail.com> writes: > >> From: Karthik Nayak <karthik.188@gmail.com> >> >> In the previous commits, we added the required base for adding symref >> commands to the '--stdin' mode provided by 'git-update-ref(1)'. Using >> them, add a new 'symref-verify' command to verify symrefs. >> >> The 'symref-verify' command allows users to verify if a provided <ref> >> contains the provided <old-target> without changing the <ref>. If >> <old-target> is not provided, the command will verify that the <ref> >> doesn't exist. Since we're checking for symbolic refs, this command will >> only work with the 'no-deref' mode. This is because any dereferenced >> symbolic ref will point to an object and not a ref and the regular >> 'verify' command can be used in such situations. > > All makes sense, but a naïve reader may find it helpful if you > explained why having "verify" command is a good idea in the first > place ("I can just do 'git symoblic-ref' to read the current value, > and see if it is what I expect"). Presumably the value of "verify" > is that you can have it in a transaction and fail other operations > in the same transaction if the symref moved from what you expected > it to point at? > I would say none of the commits drive this point, and I would go ahead and add something on these lines to each of them. I think it would add good value to readers. >> Add and use `ref_update_is_null_new_value`, a helper function which is >> used to check if there is a new_value in a reference update. The new >> value could either be a symref target `new_target` or a OID `new_oid`. >> We also add tests to test the command in both the regular stdin mode and >> also with the '-z' flag. > > This looks out of place, primarily because the helper function is > *NOT* used in this step. Without any actual user, and with the name > that says only what it checks without hinting why a caller may want > to check the condition it checks, it is hard to guess if it is a > good idea to have such a helper. > I think over the revision, its usage from this commit was removed. It makes sense to move it to a commit where its used, I'll do that. > "If a ref_update object specifies no new-oid and no new-target, it > is not about updating but just validating" is how the callers are > expected to use it, then instead of is_null_new_value that says > what it checks, something like is_verify_only that says what the > caller may want to use it for would be a more friendly name for > readers and future developers. This is true for the old-oid and old-target. That is, when they are set to null, we're validating. With the new-oid and new-target, if they're null, it usually signifies deletion. We could rename it to 'is_delete_only', but that would also need checking the 'REF_HAVE_NEW' flag. So we could ideally change it to ``` int ref_update_is_delete_only(struct ref_update *update) { return (update->flags & REF_HAVE_NEW) && !update->new_target && is_null_oid(&update->new_oid); } ``` I'm okay with making this change. >> @@ -297,11 +320,47 @@ static void parse_cmd_verify(struct ref_transaction *transaction, >> die("verify %s: extra input: %s", refname, next); >> >> if (ref_transaction_verify(transaction, refname, &old_oid, >> - update_flags, &err)) >> + NULL, update_flags, &err)) >> >> update_flags = default_flags; >> free(refname); >> strbuf_release(&err); >> } > > The only damage by this patch to parse_cmd_verify() is that > ref_transaction_verify() gained another parameter NULL, but with the > default "--diff-algorithm=myers" algorithm, it is very hard to see. > > The "--patience" algorithm does a much beter job on this hunk. > > And the following function is entirely new. > >> +static void parse_cmd_symref_verify(struct ref_transaction *transaction, >> + const char *next, const char *end) >> +{ >> + struct strbuf err = STRBUF_INIT; >> + struct object_id old_oid; >> + char *refname, *old_target; >> + >> + if (!(update_flags & REF_NO_DEREF)) >> + die("symref-verify: cannot operate with deref mode"); >> + >> + refname = parse_refname(&next); >> + if (!refname) >> + die("symref-verify: missing <ref>"); >> + >> + /* >> + * old_ref is optional, but we want to differentiate between >> + * a NULL and zero value. >> + */ >> + old_target = parse_next_refname(&next); >> + if (!old_target) >> + old_oid = *null_oid(); > > In many existing code paths, we do not do structure assignment like > this. Instead we do > > oidcpy(&old_oid, null_oid()); > > We can see an existing example in a common context in a hunk for > refs.c in this patch. > Yeah, makes sense to switch this. Will do. >> + if (*next != line_termination) >> + die("symref-verify %s: extra input: %s", refname, next); >> + >> + if (ref_transaction_verify(transaction, refname, >> + old_target ? NULL : &old_oid, >> + old_target, update_flags, &err)) >> + die("%s", err.buf); > > Are static analyzers smart enough to notice that we will not be > using old_oid uninitialized here? Just wondering. Yup, at least the clang LSP server seems to detect and not bug me about it. > Anyway. This ensures ref_transaction_verify() gets either > old_target or old_oid, but never both at the same time. The caller > to ref_transaction_verify() in the previous function passed NULL for > old_target but it always had a non-NULL old_oid so that is perfectly > fine. > >> + update_flags = default_flags; >> + free(refname); >> + free(old_target); >> + strbuf_release(&err); >> +} > >> diff --git a/refs.c b/refs.c >> index 060a31616d..0e1013b5ab 100644 >> --- a/refs.c >> +++ b/refs.c >> @@ -1217,6 +1217,8 @@ void ref_transaction_free(struct ref_transaction *transaction) >> >> for (i = 0; i < transaction->nr; i++) { >> free(transaction->updates[i]->msg); >> + free((void *)transaction->updates[i]->old_target); >> + free((void *)transaction->updates[i]->new_target); >> free(transaction->updates[i]); >> } >> free(transaction->updates); >> @@ -1247,9 +1249,13 @@ struct ref_update *ref_transaction_add_update( >> >> update->flags = flags; >> >> - if (flags & REF_HAVE_NEW) >> + if (new_target) >> + update->new_target = xstrdup(new_target); >> + if (old_target) >> + update->old_target = xstrdup(old_target); > > Presumably "update" structure, when freshly initialized, has NULL in > both of these _target members? Otherwise ref_transaction_free() > would get in trouble, so double checking. > This is a good point. My understanding was that FLEX_ALLOC_MEM should set everything to 0. >> + if (new_oid && flags & REF_HAVE_NEW) >> oidcpy(&update->new_oid, new_oid); >> - if (flags & REF_HAVE_OLD) >> + if (old_oid && flags & REF_HAVE_OLD) >> oidcpy(&update->old_oid, old_oid); > > Since we can ask to work on a symbolic ref, new_oid / old_oid can be > NULL when REF_HAVE_NEW / REF_HAVE_OLD bit is on for _target members. > > Makes me wonder if the code becomes easier to follow if the flag > bits are split into four (_NEW -> _NEW_OID + _NEW_TARGET), but let's > not worry about that for now. > The intersection of this is quite low currently, so I'm not really sure if there's added benefit. I did start that way before, but perhaps with the iterations in the last few version, maybe it makes the code simpler. >> @@ -1286,6 +1292,7 @@ int ref_transaction_update(struct ref_transaction *transaction, >> flags &= REF_TRANSACTION_UPDATE_ALLOWED_FLAGS; >> >> flags |= (new_oid ? REF_HAVE_NEW : 0) | (old_oid ? REF_HAVE_OLD : 0); >> + flags |= (new_target ? REF_HAVE_NEW : 0) | (old_target ? REF_HAVE_OLD : 0); > >> @@ -1325,14 +1332,17 @@ int ref_transaction_delete(struct ref_transaction *transaction, >> int ref_transaction_verify(struct ref_transaction *transaction, >> const char *refname, >> const struct object_id *old_oid, >> + const char *old_target, >> unsigned int flags, >> struct strbuf *err) >> { >> - if (!old_oid) >> - BUG("verify called with old_oid set to NULL"); >> + if (!old_target && !old_oid) >> + BUG("verify called with old_oid and old_target set to NULL"); > > Is it normal if you get _both_ set, or is it equally a BUG()? > The parse_*_verify() codepaths we saw earlier both made sure > only one of the two is non-NULL, and it is unclear what should > happen if both are non-NULL. > It is a bug and this is caught in `ref_transaction_add_update`. Introduced in the first patch of the series. >> + if (old_target && !(flags & REF_NO_DEREF)) >> + BUG("verify cannot operate on symrefs with deref mode"); >> return ref_transaction_update(transaction, refname, >> NULL, old_oid, >> - NULL, NULL, >> + NULL, old_target, >> flags, NULL, err); >> } > > So this queues an ref_update object whose .new_oid and .new_target > are NULL, and .old_oid and .old_target are what the caller gave us > to check. The NULLs in .new* members hopefully do not mean "delete > this thing" ;-) > So the 'new_oid' being set to zero should be the delete this thing queue. >> @@ -2349,6 +2359,12 @@ static int run_transaction_hook(struct ref_transaction *transaction, >> for (i = 0; i < transaction->nr; i++) { >> struct ref_update *update = transaction->updates[i]; >> >> + /* >> + * Skip reference transaction for symbolic refs. >> + */ >> + if (update->new_target || update->old_target) >> + continue; > > Is that a final design, or will the hooks have a chance to interfere? > The last patch adds hook support. >> diff --git a/refs/files-backend.c b/refs/files-backend.c >> index 2420dac2aa..53197fa3af 100644 >> --- a/refs/files-backend.c >> +++ b/refs/files-backend.c >> @@ -2425,6 +2425,37 @@ static const char *original_update_refname(struct ref_update *update) >> return update->refname; >> } >> >> +/* >> + * Check whether the REF_HAVE_OLD and old_target values stored in >> + * update are consistent with ref, which is the symbolic reference's >> + * current value. If everything is OK, return 0; otherwise, write an >> + * error message to err and return -1. >> + */ >> +static int check_old_target(struct ref_update *update, char *ref, >> + struct strbuf *err) >> +{ >> + if (!(update->flags & REF_HAVE_OLD) || >> + !strcmp(update->old_target, ref)) >> + return 0; > > Earlier on the assignment side for "update" structure we saw above, > the guard was (old_target && flags & REF_HAVE_OLD), but here we > assume old_target is valid, which feels a bit asymmetric. > > Yes, I can see that the caller does not call us when !old_target, > but still... Perhaps > > if ((update->flags & REF_HAVE_OLD) && !update->old_target) > BUG(...); > I will add something like this. > or something? Or alternatively, perhaps !!update->old_target should > be the only thing we should check and ignore REF_HAVE_OLD bit? I am > not sure, but it smells like that the non-NULL-ness of old_target is > the only thing that matters (if it is not NULL, very early in the > control flow somebody would have set REF_HAVE_OLD bit to flags, no?). > > It brings me back to my earlier question. Does REF_HAVE_OLD bit > serve a useful purpose in this code? > I checked and it doesn't, it can be removed from usage in this code. Will cleanup this part. >> + if (!strcmp(update->old_target, "")) >> + strbuf_addf(err, "cannot lock ref '%s': " >> + "reference already exists", >> + original_update_refname(update)); >> + else if (!strcmp(ref, "")) >> + strbuf_addf(err, "cannot lock ref '%s': " >> + "reference is missing but expected %s", >> + original_update_refname(update), >> + update->old_target); > > So... for old_target and ref, an empty string is a special value? > How? Shouldn't that be documented in the comment before the > function? > >> + else >> + strbuf_addf(err, "cannot lock ref '%s': " >> + "is at %s but expected %s", >> + original_update_refname(update), >> + ref, update->old_target); >> + >> + return -1; >> +} >> + >> /* >> * Check whether the REF_HAVE_OLD and old_oid values stored in update >> * are consistent with oid, which is the reference's current value. If >> @@ -2528,6 +2559,18 @@ static int lock_ref_for_update(struct files_ref_store *refs, >> ret = TRANSACTION_GENERIC_ERROR; >> goto out; >> } >> + } >> + >> + /* >> + * For symref verification, we need to check the reference value >> + * rather than the oid. If we're dealing with regular refs or we're >> + * verifying a dereferenced symref, we then check the oid. >> + */ >> + if (update->old_target) { >> + if (check_old_target(update, referent.buf, err)) { >> + ret = TRANSACTION_GENERIC_ERROR; >> + goto out; >> + } > > We come here only when update->type has REF_ISSYMREF bit on (we > learned that value by calling lock_raw_ref()), and know referent.buf > has the current "target" value. That is consumed as "ref" parameter > to check_old_target() we just saw. OK. > >> diff --git a/refs/reftable-backend.c b/refs/reftable-backend.c >> index 6104471199..a2474245aa 100644 >> --- a/refs/reftable-backend.c >> +++ b/refs/reftable-backend.c >> @@ -938,7 +938,26 @@ static int reftable_be_transaction_prepare(struct ref_store *ref_store, >> * individual refs. But the error messages match what the files >> * backend returns, which keeps our tests happy. >> */ >> - if (u->flags & REF_HAVE_OLD && !oideq(¤t_oid, &u->old_oid)) { >> + if ((u->flags & REF_HAVE_OLD) && u->old_target) { >> + if (strcmp(referent.buf, u->old_target)) { >> + if (!strcmp(u->old_target, "")) >> + strbuf_addf(err, "verifying symref target: '%s': " >> + "provided target is empty", >> + original_update_refname(u)); >> + else if (!strcmp(referent.buf, "")) >> + strbuf_addf(err, "verifying symref target: '%s': " >> + "reference is missing but expected %s", >> + original_update_refname(u), >> + u->old_target); >> + else >> + strbuf_addf(err, "verifying symref target: '%s': " >> + "is at %s but expected %s", >> + original_update_refname(u), >> + referent.buf, u->old_target); >> + ret = -1; >> + goto done; >> + } > > Again, the puzzling "empty string"s are handled here. For here and above, this too is dead code and no longer needed, old_target being empty string is left over code from before we decided to use zero_oid for deleting. I'll remove it. Thanks.
diff --git a/Documentation/git-update-ref.txt b/Documentation/git-update-ref.txt index 374a2ebd2b..9fe78b3501 100644 --- a/Documentation/git-update-ref.txt +++ b/Documentation/git-update-ref.txt @@ -65,6 +65,7 @@ performs all modifications together. Specify commands of the form: create SP <ref> SP <new-oid> LF delete SP <ref> [SP <old-oid>] LF verify SP <ref> [SP <old-oid>] LF + symref-verify SP <ref> [SP <old-target>] LF option SP <opt> LF start LF prepare LF @@ -86,6 +87,7 @@ quoting: create SP <ref> NUL <new-oid> NUL delete SP <ref> NUL [<old-oid>] NUL verify SP <ref> NUL [<old-oid>] NUL + symref-verify SP <ref> [NUL <old-target>] NUL option SP <opt> NUL start NUL prepare NUL @@ -117,6 +119,11 @@ verify:: Verify <ref> against <old-oid> but do not change it. If <old-oid> is zero or missing, the ref must not exist. +symref-verify:: + Verify symbolic <ref> against <old-target> but do not change it. + If <old-target> is missing, the ref must not exist. Can only be + used in `no-deref` mode. + option:: Modify the behavior of the next command naming a <ref>. The only valid option is `no-deref` to avoid dereferencing diff --git a/builtin/update-ref.c b/builtin/update-ref.c index 21fdbf6ac8..419b28169b 100644 --- a/builtin/update-ref.c +++ b/builtin/update-ref.c @@ -76,6 +76,29 @@ static char *parse_refname(const char **next) return strbuf_detach(&ref, NULL); } +/* + * Wrapper around parse_refname which skips the next delimiter. + */ +static char *parse_next_refname(const char **next) +{ + if (line_termination) { + /* Without -z, consume SP and use next argument */ + if (!**next || **next == line_termination) + return NULL; + if (**next != ' ') + die("expected SP but got: %s", *next); + } else { + /* With -z, read the next NUL-terminated line */ + if (**next) + return NULL; + } + /* Skip the delimiter */ + (*next)++; + + return parse_refname(next); +} + + /* * The value being parsed is <old-oid> (as opposed to <new-oid>; the * difference affects which error messages are generated): @@ -297,11 +320,47 @@ static void parse_cmd_verify(struct ref_transaction *transaction, die("verify %s: extra input: %s", refname, next); if (ref_transaction_verify(transaction, refname, &old_oid, - update_flags, &err)) + NULL, update_flags, &err)) + die("%s", err.buf); + + update_flags = default_flags; + free(refname); + strbuf_release(&err); +} + +static void parse_cmd_symref_verify(struct ref_transaction *transaction, + const char *next, const char *end) +{ + struct strbuf err = STRBUF_INIT; + struct object_id old_oid; + char *refname, *old_target; + + if (!(update_flags & REF_NO_DEREF)) + die("symref-verify: cannot operate with deref mode"); + + refname = parse_refname(&next); + if (!refname) + die("symref-verify: missing <ref>"); + + /* + * old_ref is optional, but we want to differentiate between + * a NULL and zero value. + */ + old_target = parse_next_refname(&next); + if (!old_target) + old_oid = *null_oid(); + + if (*next != line_termination) + die("symref-verify %s: extra input: %s", refname, next); + + if (ref_transaction_verify(transaction, refname, + old_target ? NULL : &old_oid, + old_target, update_flags, &err)) die("%s", err.buf); update_flags = default_flags; free(refname); + free(old_target); strbuf_release(&err); } @@ -380,15 +439,16 @@ static const struct parse_cmd { unsigned args; enum update_refs_state state; } command[] = { - { "update", parse_cmd_update, 3, UPDATE_REFS_OPEN }, - { "create", parse_cmd_create, 2, UPDATE_REFS_OPEN }, - { "delete", parse_cmd_delete, 2, UPDATE_REFS_OPEN }, - { "verify", parse_cmd_verify, 2, UPDATE_REFS_OPEN }, - { "option", parse_cmd_option, 1, UPDATE_REFS_OPEN }, - { "start", parse_cmd_start, 0, UPDATE_REFS_STARTED }, - { "prepare", parse_cmd_prepare, 0, UPDATE_REFS_PREPARED }, - { "abort", parse_cmd_abort, 0, UPDATE_REFS_CLOSED }, - { "commit", parse_cmd_commit, 0, UPDATE_REFS_CLOSED }, + { "update", parse_cmd_update, 3, UPDATE_REFS_OPEN }, + { "create", parse_cmd_create, 2, UPDATE_REFS_OPEN }, + { "delete", parse_cmd_delete, 2, UPDATE_REFS_OPEN }, + { "verify", parse_cmd_verify, 2, UPDATE_REFS_OPEN }, + { "symref-verify", parse_cmd_symref_verify, 2, UPDATE_REFS_OPEN }, + { "option", parse_cmd_option, 1, UPDATE_REFS_OPEN }, + { "start", parse_cmd_start, 0, UPDATE_REFS_STARTED }, + { "prepare", parse_cmd_prepare, 0, UPDATE_REFS_PREPARED }, + { "abort", parse_cmd_abort, 0, UPDATE_REFS_CLOSED }, + { "commit", parse_cmd_commit, 0, UPDATE_REFS_CLOSED }, }; static void update_refs_stdin(void) diff --git a/refs.c b/refs.c index 060a31616d..0e1013b5ab 100644 --- a/refs.c +++ b/refs.c @@ -1217,6 +1217,8 @@ void ref_transaction_free(struct ref_transaction *transaction) for (i = 0; i < transaction->nr; i++) { free(transaction->updates[i]->msg); + free((void *)transaction->updates[i]->old_target); + free((void *)transaction->updates[i]->new_target); free(transaction->updates[i]); } free(transaction->updates); @@ -1247,9 +1249,13 @@ struct ref_update *ref_transaction_add_update( update->flags = flags; - if (flags & REF_HAVE_NEW) + if (new_target) + update->new_target = xstrdup(new_target); + if (old_target) + update->old_target = xstrdup(old_target); + if (new_oid && flags & REF_HAVE_NEW) oidcpy(&update->new_oid, new_oid); - if (flags & REF_HAVE_OLD) + if (old_oid && flags & REF_HAVE_OLD) oidcpy(&update->old_oid, old_oid); update->msg = normalize_reflog_message(msg); return update; @@ -1286,6 +1292,7 @@ int ref_transaction_update(struct ref_transaction *transaction, flags &= REF_TRANSACTION_UPDATE_ALLOWED_FLAGS; flags |= (new_oid ? REF_HAVE_NEW : 0) | (old_oid ? REF_HAVE_OLD : 0); + flags |= (new_target ? REF_HAVE_NEW : 0) | (old_target ? REF_HAVE_OLD : 0); ref_transaction_add_update(transaction, refname, flags, new_oid, old_oid, new_target, @@ -1325,14 +1332,17 @@ int ref_transaction_delete(struct ref_transaction *transaction, int ref_transaction_verify(struct ref_transaction *transaction, const char *refname, const struct object_id *old_oid, + const char *old_target, unsigned int flags, struct strbuf *err) { - if (!old_oid) - BUG("verify called with old_oid set to NULL"); + if (!old_target && !old_oid) + BUG("verify called with old_oid and old_target set to NULL"); + if (old_target && !(flags & REF_NO_DEREF)) + BUG("verify cannot operate on symrefs with deref mode"); return ref_transaction_update(transaction, refname, NULL, old_oid, - NULL, NULL, + NULL, old_target, flags, NULL, err); } @@ -2349,6 +2359,12 @@ static int run_transaction_hook(struct ref_transaction *transaction, for (i = 0; i < transaction->nr; i++) { struct ref_update *update = transaction->updates[i]; + /* + * Skip reference transaction for symbolic refs. + */ + if (update->new_target || update->old_target) + continue; + strbuf_reset(&buf); strbuf_addf(&buf, "%s %s %s\n", oid_to_hex(&update->old_oid), @@ -2802,3 +2818,7 @@ int copy_existing_ref(const char *oldref, const char *newref, const char *logmsg { return refs_copy_existing_ref(get_main_ref_store(the_repository), oldref, newref, logmsg); } + +int ref_update_is_null_new_value(struct ref_update *update) { + return !update->new_target && is_null_oid(&update->new_oid); +} diff --git a/refs.h b/refs.h index c792e13a64..27b9aeaf54 100644 --- a/refs.h +++ b/refs.h @@ -780,6 +780,7 @@ int ref_transaction_delete(struct ref_transaction *transaction, int ref_transaction_verify(struct ref_transaction *transaction, const char *refname, const struct object_id *old_oid, + const char *old_target, unsigned int flags, struct strbuf *err); diff --git a/refs/files-backend.c b/refs/files-backend.c index 2420dac2aa..53197fa3af 100644 --- a/refs/files-backend.c +++ b/refs/files-backend.c @@ -2425,6 +2425,37 @@ static const char *original_update_refname(struct ref_update *update) return update->refname; } +/* + * Check whether the REF_HAVE_OLD and old_target values stored in + * update are consistent with ref, which is the symbolic reference's + * current value. If everything is OK, return 0; otherwise, write an + * error message to err and return -1. + */ +static int check_old_target(struct ref_update *update, char *ref, + struct strbuf *err) +{ + if (!(update->flags & REF_HAVE_OLD) || + !strcmp(update->old_target, ref)) + return 0; + + if (!strcmp(update->old_target, "")) + strbuf_addf(err, "cannot lock ref '%s': " + "reference already exists", + original_update_refname(update)); + else if (!strcmp(ref, "")) + strbuf_addf(err, "cannot lock ref '%s': " + "reference is missing but expected %s", + original_update_refname(update), + update->old_target); + else + strbuf_addf(err, "cannot lock ref '%s': " + "is at %s but expected %s", + original_update_refname(update), + ref, update->old_target); + + return -1; +} + /* * Check whether the REF_HAVE_OLD and old_oid values stored in update * are consistent with oid, which is the reference's current value. If @@ -2528,6 +2559,18 @@ static int lock_ref_for_update(struct files_ref_store *refs, ret = TRANSACTION_GENERIC_ERROR; goto out; } + } + + /* + * For symref verification, we need to check the reference value + * rather than the oid. If we're dealing with regular refs or we're + * verifying a dereferenced symref, we then check the oid. + */ + if (update->old_target) { + if (check_old_target(update, referent.buf, err)) { + ret = TRANSACTION_GENERIC_ERROR; + goto out; + } } else if (check_old_oid(update, &lock->old_oid, err)) { ret = TRANSACTION_GENERIC_ERROR; goto out; diff --git a/refs/refs-internal.h b/refs/refs-internal.h index 3040d4797c..23e65f65e8 100644 --- a/refs/refs-internal.h +++ b/refs/refs-internal.h @@ -748,4 +748,11 @@ void base_ref_store_init(struct ref_store *refs, struct repository *repo, */ struct ref_store *maybe_debug_wrap_ref_store(const char *gitdir, struct ref_store *store); +/* + * Helper function to check if the new value is null, this + * takes into consideration that the update could be a regular + * ref or a symbolic ref. + */ +int ref_update_is_null_new_value(struct ref_update *update); + #endif /* REFS_REFS_INTERNAL_H */ diff --git a/refs/reftable-backend.c b/refs/reftable-backend.c index 6104471199..a2474245aa 100644 --- a/refs/reftable-backend.c +++ b/refs/reftable-backend.c @@ -938,7 +938,26 @@ static int reftable_be_transaction_prepare(struct ref_store *ref_store, * individual refs. But the error messages match what the files * backend returns, which keeps our tests happy. */ - if (u->flags & REF_HAVE_OLD && !oideq(¤t_oid, &u->old_oid)) { + if ((u->flags & REF_HAVE_OLD) && u->old_target) { + if (strcmp(referent.buf, u->old_target)) { + if (!strcmp(u->old_target, "")) + strbuf_addf(err, "verifying symref target: '%s': " + "provided target is empty", + original_update_refname(u)); + else if (!strcmp(referent.buf, "")) + strbuf_addf(err, "verifying symref target: '%s': " + "reference is missing but expected %s", + original_update_refname(u), + u->old_target); + else + strbuf_addf(err, "verifying symref target: '%s': " + "is at %s but expected %s", + original_update_refname(u), + referent.buf, u->old_target); + ret = -1; + goto done; + } + } else if (u->flags & REF_HAVE_OLD && !oideq(¤t_oid, &u->old_oid)) { if (is_null_oid(&u->old_oid)) strbuf_addf(err, _("cannot lock ref '%s': " "reference already exists"), diff --git a/t/t1400-update-ref.sh b/t/t1400-update-ref.sh index ec3443cc87..34b29eeac8 100755 --- a/t/t1400-update-ref.sh +++ b/t/t1400-update-ref.sh @@ -890,17 +890,23 @@ test_expect_success 'stdin update/create/verify combination works' ' ' test_expect_success 'stdin verify succeeds for correct value' ' + test-tool ref-store main for-each-reflog-ent $m >before && git rev-parse $m >expect && echo "verify $m $m" >stdin && git update-ref --stdin <stdin && git rev-parse $m >actual && - test_cmp expect actual + test_cmp expect actual && + test-tool ref-store main for-each-reflog-ent $m >after && + test_cmp before after ' test_expect_success 'stdin verify succeeds for missing reference' ' + test-tool ref-store main for-each-reflog-ent $m >before && echo "verify refs/heads/missing $Z" >stdin && git update-ref --stdin <stdin && - test_must_fail git rev-parse --verify -q refs/heads/missing + test_must_fail git rev-parse --verify -q refs/heads/missing && + test-tool ref-store main for-each-reflog-ent $m >after && + test_cmp before after ' test_expect_success 'stdin verify treats no value as missing' ' @@ -1641,4 +1647,87 @@ test_expect_success PIPE 'transaction flushes status updates' ' test_cmp expected actual ' +create_stdin_buf () { + if test "$1" = "-z" + then + shift + printf "$F" "$@" >stdin + else + echo "$@" >stdin + fi +} + +for type in "" "-z" +do + + test_expect_success "stdin ${type} symref-verify fails without --no-deref" ' + git symbolic-ref refs/heads/symref $a && + create_stdin_buf ${type} "symref-verify refs/heads/symref" "$a" && + test_must_fail git update-ref --stdin ${type} <stdin 2>err && + grep "fatal: symref-verify: cannot operate with deref mode" err + ' + + test_expect_success "stdin ${type} symref-verify fails with too many arguments" ' + create_stdin_buf ${type} "symref-verify refs/heads/symref" "$a" "$a" && + test_must_fail git update-ref --stdin ${type} --no-deref <stdin 2>err && + if test "$type" = "-z" + then + grep "fatal: unknown command: $a" err + else + grep "fatal: symref-verify refs/heads/symref: extra input: $a" err + fi + ' + + test_expect_success "stdin ${type} symref-verify succeeds for correct value" ' + git symbolic-ref refs/heads/symref >expect && + test-tool ref-store main for-each-reflog-ent refs/heads/symref >before && + create_stdin_buf ${type} "symref-verify refs/heads/symref" "$a" && + git update-ref --stdin ${type} --no-deref <stdin && + git symbolic-ref refs/heads/symref >actual && + test_cmp expect actual && + test-tool ref-store main for-each-reflog-ent refs/heads/symref >after && + test_cmp before after + ' + + test_expect_success "stdin ${type} symref-verify no value is treated as zero value" ' + git symbolic-ref refs/heads/symref >expect && + create_stdin_buf ${type} "symref-verify refs/heads/symref" "" && + test_must_fail git update-ref --stdin ${type} --no-deref <stdin + ' + + test_expect_success "stdin ${type} symref-verify succeeds for dangling reference" ' + test_when_finished "git symbolic-ref -d refs/heads/symref2" && + test_must_fail git symbolic-ref refs/heads/nonexistent && + git symbolic-ref refs/heads/symref2 refs/heads/nonexistent && + create_stdin_buf ${type} "symref-verify refs/heads/symref2" "refs/heads/nonexistent" && + git update-ref --stdin ${type} --no-deref <stdin + ' + + test_expect_success "stdin ${type} symref-verify succeeds for missing reference" ' + test-tool ref-store main for-each-reflog-ent refs/heads/symref >before && + create_stdin_buf ${type} "symref-verify refs/heads/missing" "$Z" && + git update-ref --stdin ${type} --no-deref <stdin && + test_must_fail git rev-parse --verify -q refs/heads/missing && + test-tool ref-store main for-each-reflog-ent refs/heads/symref >after && + test_cmp before after + ' + + test_expect_success "stdin ${type} symref-verify fails for wrong value" ' + git symbolic-ref refs/heads/symref >expect && + create_stdin_buf ${type} "symref-verify refs/heads/symref" "$b" && + test_must_fail git update-ref --stdin ${type} --no-deref <stdin && + git symbolic-ref refs/heads/symref >actual && + test_cmp expect actual + ' + + test_expect_success "stdin ${type} symref-verify fails for mistaken null value" ' + git symbolic-ref refs/heads/symref >expect && + create_stdin_buf ${type} "symref-verify refs/heads/symref" "$Z" && + test_must_fail git update-ref --stdin ${type} --no-deref <stdin && + git symbolic-ref refs/heads/symref >actual && + test_cmp expect actual + ' + +done + test_done