diff options
Diffstat (limited to 'CHANGES')
| -rw-r--r-- | CHANGES | 333 |
1 files changed, 326 insertions, 7 deletions
@@ -1,3 +1,315 @@ +Changes from V8.6beta1 to V8.6 +============================== + +Kernel + +- Fixed critical bug #5248 in VM long multiplication on 32-bit + architectures. Was there only since 8.6beta1, so no stable release impacted. + +Other bug fixes in universes, type class shelving,... + +Changes from V8.5 to V8.6beta1 +============================== + +Kernel + +- A new, faster state-of-the-art universe constraint checker. + +Specification language + +- Giving implicit arguments explicitly to a constant with multiple + choices of implicit arguments does not break any more insertion of + further maximal implicit arguments. +- Ability to put any pattern in binders, prefixed by quote, e.g. + "fun '(a,b) => ...", "λ '(a,(b,c)), ...", "Definition foo '(x,y) := ...". + It expands into a "let 'pattern := ..." + +Tactics + +- Flag "Bracketing Last Introduction Pattern" is now on by default. +- Flag "Regular Subst Tactic" is now on by default: it respects the + initial order of hypothesis, it contracts cycles, it unfolds no + local definitions (common source of incompatibilities, fixable by + "Unset Regular Subst Tactic"). +- New flag "Refolding Reduction", now disabled by default, which turns + on refolding of constants/fixpoints (as in cbn) during the reductions + done during type inference and tactic retyping. Can be extremely + expensive. When set off, this recovers the 8.4 behaviour of unification + and type inference. Potential source of incompatibility with 8.5 developments + (the option is set on in Compat/Coq85.v). +- New flag "Shrink Abstract" that minimalizes proofs generated by the abstract + tactical w.r.t. variables appearing in the body of the proof. + On by default and deprecated. Minor source of incompatibility + for code relying on the precise arguments of abstracted proofs. +- Serious bugs are fixed in tactic "double induction" (source of + incompatibilities as soon as the inductive types have dependencies in + the type of their constructors; "double induction" remains however + deprecated). +- In introduction patterns of the form (pat1,...,patn), n should match + the exact number of hypotheses introduced (except for local definitions + for which pattern can be omitted, as in regular pattern-matching). +- Tactic scopes in Ltac like constr: and ltac: now require parentheses around + their argument. +- Every generic argument type declares a tactic scope of the form "name:(...)" + where name is the name of the argument. This generalizes the constr: and ltac: + instances. +- When in strict mode (i.e. in a Ltac definition), if the "intro" tactic is + given a free identifier, it is not bound in subsequent tactics anymore. + In order to introduce a binding, use e.g. the "fresh" primitive instead + (potential source of incompatibilities). +- New tactics is_ind, is_const, is_proj, is_constructor for use in Ltac. +- New goal selectors. Sets of goals can be selected by listing integers + ranges. Example: "1,4-7,24: tac" focuses "tac" on goals 1,4,5,6,7,24. +- For uniformity with "destruct"/"induction" and for a more natural + behavior, "injection" can now work in place by activating option + "Structural Injection". In this case, hypotheses are also put in the + context in the natural left-to-right order and the hypothesis on + which injection applies is cleared. +- Tactic "contradiction" (hence "easy") now also solve goals with + hypotheses of the form "~True" or "t<>t" (possible source of + incompatibilities because of more successes in automation, but + generally a more intuitive strategy). +- Option "Injection On Proofs" was renamed "Keep Proof Equalities". When + enabled, injection and inversion do not drop equalities between objects + in Prop. Still disabled by default. +- New tactics "notypeclasses refine" and "simple notypeclasses refine" that + disallow typeclass resolution when typechecking their argument, for use + in typeclass hints. +- Integration of LtacProf, a profiler for Ltac. +- Reduction tactics now accept more fine-grained flags: iota is now a shorthand + for the new flags match, fix and cofix. +- The ssreflect subterm selection algorithm is now accessible to tactic writers + through the ssrmatching plugin. +- When used as an argument of an ltac function, "auto" without "with" + nor "using" clause now correctly uses only the core hint database by + default. + +Hints + +- Revised the syntax of [Hint Cut] to follow standard notation for regexps. +- Hint Mode now accepts "!" which means that the mode matches only if the + argument's head is not an evar (it goes under applications, casts, and + scrutinees of matches and projections). +- Hints can now take an optional user-given pattern, used only by + [typeclasses eauto] with the [Filtered Unification] option on. + +Typeclasses + +- Many new options and new engine based on the proof monad. The + [typeclasses eauto] tactic is now a multi-goal, multi-success tactic. + See reference manual for more information. It is planned to + replace auto and eauto in the following version. The 8.5 resolution + engine is still available to help solve compatibility issues. + +Program + +- The "Shrink Obligations" flag now applies to all obligations, not only + those solved by the automatic tactic. +- "Shrink Obligations" is on by default and deprecated. Minor source of + incompatibility for code relying on the precise arguments of + obligations. + +Notations + +- "Bind Scope" can once again bind "Funclass" and "Sortclass". + +General infrastructure + +- New configurable warning system which can be controlled with the vernacular + command "Set Warnings", or, under coqc/coqtop, with the flag "-w". In + particular, the default is now that warnings are printed by coqc. +- In asynchronous mode, Coq is now capable of recovering from errors and + continue processing the document. + +Tools + +- coqc accepts a -o option to specify the output file name +- coqtop accepts --print-version to print Coq and OCaml versions in + easy to parse format +- Setting [Printing Dependent Evars Line] can be unset to disable the + computation associated with printing the "dependent evars: " line in + -emacs mode +- Removed the -verbose-compat-notations flag and the corresponding Set + Verbose Compat vernacular, since these warnings can now be silenced or + turned into errors using "-w". + +XML protocol + +- message format has changed, see dev/doc/changes.txt for more details. + +Many bug fixes, minor changes and documentation improvements are not mentioned +here. + +Changes from V8.5pl2 to V8.5pl3 +=============================== + +Critical bugfix + +- #4876: Guard checker incompleteness when using primitive projections + +Other bugfixes + +- #4780: Induction with universe polymorphism on was creating ill-typed terms. +- #4673: regression in setoid_rewrite, unfolding let-ins for type unification. +- #4754: Regression in setoid_rewrite, allow postponed unification problems to remain. +- #4769: Anomaly with universe polymorphic schemes defined inside sections. +- #3886: Program: duplicate obligations of mutual fixpoints. +- #4994: Documentation typo. +- #5008: Use the "md5" command on OpenBSD. +- #5007: Do not assume the "TERM" environment variable is always set. +- #4606: Output a break before a list only if there was an empty line. +- #5001: metas not cleaned properly in clenv_refine_in. +- #2336: incorrect glob data for module symbols (bug #2336). +- #4832: Remove extraneous dot in error message. +- Anomaly in printing a unification error message. +- #4947: Options which take string arguments are not backwards compatible. +- #4156: micromega cache files are now hidden files. +- #4871: interrupting par:abstract kills coqtop. +- #5043: [Admitted] lemmas pick up section variables. +- Fix name of internal refine ("simple refine"). +- #5062: probably a typo in Strict Proofs mode. +- #5065: Anomaly: Not a proof by induction. +- Restore native compiler optimizations, they were disabled since 8.5! +- #5077: failure on typing a fixpoint with evars in its type. +- Fix recursive notation bug. +- #5095: non relevant too strict test in let-in abstraction. +- Ensuring that the evar name is preserved by "rename". +- #4887: confusion between using and with in documentation of firstorder. +- Bug in subst with let-ins. +- #4762: eauto weaker than auto. +- Remove if_then_else (was buggy). Use tryif instead. +- #4970: confusion between special "{" and non special "{{" in notations. +- #4529: primitive projections unfolding. +- #4416: Incorrect "Error: Incorrect number of goals". +- #4863: abstract in typeclass hint fails. +- #5123: unshelve can impact typeclass resolution +- Fix a collision about the meta-variable ".." in recursive notations. +- Fix printing of info_auto. +- #3209: Not_found due to an occur-check cycle. +- #5097: status of evars refined by "clear" in ltac: closed wrt evars. +- #5150: Missing dependency of the test-suite subsystems in prerequisite. +- Fix a bug in error printing of unif constraints +- #3941: Do not stop propagation of signals when Coq is busy. +- #4822: Incorrect assertion in cbn. +- #3479 parsing of "{" and "}" when a keyword starts with "{" or "}". +- #5127: Memory corruption with the VM. +- #5102: bullets parsing broken by calls to parse_entry. + +Various documentation improvements + + +Changes from V8.5pl1 to V8.5pl2 +=============================== + +Critical bugfix +- Checksums of .vo files dependencies were not correctly checked. +- Unicode-to-ASCII translation was not injective, leading in a soundness bug in + the native compiler. + +Other bugfixes + +- #4097: more efficient occur-check in presence of primitive projections +- #4398: type_scope used consistently in "match goal". +- #4450: eauto does not work with polymorphic lemmas +- #4677: fix alpha-conversion in notations needing eta-expansion. +- Fully preserve initial order of hypotheses in "Regular Subst Tactic" mode. +- #4644: a regression in unification. +- #4725: Function (Error: Conversion test raised an anomaly) and Program + (Error: Cannot infer this placeholder of type) +- #4747: Problem building Coq 8.5pl1 with OCaml 4.03.0: Fatal warnings +- #4752: CoqIDE crash on files not ended by ".v". +- #4777: printing inefficiency with implicit arguments +- #4818: "Admitted" fails due to undefined universe anomaly after calling + "destruct" +- #4823: remote counter: avoid thread race on sockets +- #4841: -verbose flag changed semantics in 8.5, is much harder to use +- #4851: [nsatz] cannot handle duplicated hypotheses +- #4858: Anomaly: Uncaught exception Failure("hd"). Please report. in variant + of nsatz +- #4880: [nsatz_compute] generates invalid certificates if given redundant + hypotheses +- #4881: synchronizing "Declare Implicit Tactic" with backtrack. +- #4882: anomaly with Declare Implicit Tactic on hole of type with evars +- Fix use of "Declare Implicit Tactic" in refine. + triggered by CoqIDE +- #4069, #4718: congruence fails when universes are involved. + +Universes +- Disallow silently dropping universe instances applied to variables + (forward compatible) +- Allow explicit universe instances on notations, when they can apply + to the head reference of their expansion. + +Build infrastructure +- New update on how to find camlp5 binary and library at configure time. + +Changes from V8.5 to V8.5pl1 +============================ + +Critical bugfix +- The subterm relation for the guard condition was incorrectly defined on + primitive projections (#4588) + +Plugin development tools +- add a .merlin target to the makefile + +Various performance improvements (time, space used by .vo files) + +Other bugfixes + +- Fix order of arguments to Big.compare_case in ExtrOcamlZBigInt.v +- Added compatibility coercions from Specif.v which were present in Coq 8.4. +- Fixing a source of inefficiency and an artificial dependency in the printer in the congruence tactic. +- Allow to unset the refinement mode of Instance in ML +- Fixing an incorrect use of prod_appvect on a term which was not a product in setoid_rewrite. +- Add -compat 8.4 econstructor tactics, and tests +- Add compatibility Nonrecursive Elimination Schemes +- Fixing the "No applicable tactic" non informative error message regression on apply. +- Univs: fix get_current_context (bug #4603, part I) +- Fix a bug in Program coercion code +- Fix handling of arity of definitional classes. +- #4630: Some tactics are 20x slower in 8.5 than 8.4. +- #4627: records with no declared arity can be template polymorphic. +- #4623: set tactic too weak with universes (regression) +- Fix incorrect behavior of CS resolution +- #4591: Uncaught exception in directory browsing. +- CoqIDE is more resilient to initialization errors. +- #4614: "Fully check the document" is uninterruptable. +- Try eta-expansion of records only on non-recursive ones +- Fix bug when a sort is ascribed to a Record +- Primitive projections: protect kernel from erroneous definitions. +- Fixed bug #4533 with previous Keyed Unification commit +- Win: kill unreliable hence do not waitpid after kill -9 (Close #4369) +- Fix strategy of Keyed Unification +- #4608: Anomaly "output_value: abstract value (outside heap)". +- #4607: do not read native code files if native compiler was disabled. +- #4105: poor escaping in the protocol between CoqIDE and coqtop. +- #4596: [rewrite] broke in the past few weeks. +- #4533 (partial): respect declared global transparency of projections in unification.ml +- #4544: Backtrack on using full betaiota reduction during keyed unification. +- #4540: CoqIDE bottom progress bar does not update. +- Fix regression from 8.4 in reflexivity +- #4580: [Set Refine Instance Mode] also used for Program Instance. +- #4582: cannot override notation [ x ]. MAY CREATE INCOMPATIBILITIES, see #4683. +- STM: Print/Extraction have to be skipped if -quick +- #4542: CoqIDE: STOP button also stops workers +- STM: classify some variants of Instance as regular `Fork nodes. +- #4574: Anomaly: Uncaught exception Invalid_argument("splay_arity"). +- Do not give a name to anonymous evars anymore. See bug #4547. +- STM: always stock in vio files the first node (state) of a proof +- STM: not delegate proofs that contain Vernac(Module|Require|Import), #4530 +- Don't fail fatally if PATH is not set. +- #4537: Coq 8.5 is slower in typeclass resolution. +- #4522: Incorrect "Warning..." on windows. +- #4373: coqdep does not know about .vio files. +- #3826: "Incompatible module types" is uninformative. +- #4495: Failed assertion in metasyntax.ml. +- #4511: evar tactic can create non-typed evars. +- #4503: mixing universe polymorphic and monomorphic variables and definitions in sections is unsupported. +- #4519: oops, global shadowed local universe level bindings. +- #4506: Anomaly: File "pretyping/indrec.ml", line 169, characters 14-20: Assertion failed. +- #4548: Coqide crashes when going back one command + Changes from V8.5beta3 to V8.5 ============================== @@ -63,6 +375,13 @@ Tactics "intros" automatically complete the introduction of its subcomponents, as the the disjunctive-conjunctive introduction patterns in non-terminal position already do. +- New flag "Shrink Abstract" that minimalizes proofs generated by the abstract + tactical w.r.t. variables appearing in the body of the proof. + +Program + +- The "Shrink Obligations" flag now applies to all obligations, not only those +solved by the automatic tactic. - Importing Program no longer overrides the "exists" tactic (potential source of incompatibilities). - Hints costs are now correctly taken into account (potential source of @@ -353,8 +672,8 @@ Tactics - When given a reference as argument, simpl, vm_compute and native_compute now strictly interpret it as the head of a pattern starting with this reference. -- The "change p with c" tactic semantics changed, now type-checking - "c" at each matching occurrence "t" of the pattern "p", and +- The "change p with c" tactic semantics changed, now type-checking + "c" at each matching occurrence "t" of the pattern "p", and converting "t" with "c". - Now "appcontext" and "context" behave the same. The old buggy behavior of "context" can be retrieved at parse time by setting the @@ -407,6 +726,9 @@ Tactics - Behavior of introduction patterns -> and <- made more uniform (hypothesis is cleared, rewrite in hypotheses and conclusion and erasing the variable when rewriting a variable). +- New experimental option "Set Standard Proposition Elimination Names" + so that case analysis or induction on schemes in Type containing + propositions now produces "H"-based names. - Tactics from plugins are now active only when the corresponding module is imported (source of incompatibilities, solvable by adding an "Import"; in the particular case of Omega, use "Require Import OmegaTactic"). @@ -818,7 +1140,7 @@ Extraction instead of accessing their body, they are now considered as axioms. The previous behaviour can be reactivated via the option "Set Extraction AccessOpaque". -- The pretty-printer for Haskell now produces layout-independant code +- The pretty-printer for Haskell now produces layout-independent code - A new command "Separate Extraction cst1 cst2 ..." that mixes a minimal extracted environment a la "Recursive Extraction" and the production of several files (one per coq source) a la "Extraction Library" @@ -945,9 +1267,6 @@ Other tactics clears (resp. reverts) H and all the hypotheses that depend on H. - Ltac's pattern-matching now supports matching metavariables that depend on variables bound upwards in the pattern. -- New experimental option "Set Standard Proposition Elimination Names" - so that case analysis or induction on schemes in Type containing - propositions now produces "H"-based names. Tactic definitions @@ -1506,7 +1825,7 @@ Tactics Moreover, romega now has a variant "romega with *" that can be also used on non-Z goals (nat, N, positive) via a call to a translation tactic named zify (its purpose is to Z-ify your goal...). This zify may also be used - independantly of romega. + independently of romega. - Tactic "remember" now supports an "in" clause to remember only selected occurrences of a term. - Tactic "pose proof" supports name overwriting in case of specialization of an |