| Age | Commit message (Collapse) | Author |
|---|
|
|
|
|
|
The hypinfo cache was actually always set to None, so that there was no need
to try to preserve it if it was set to an actual value.
|
|
|
|
|
|
|
|
|
|
The command [Redirect "filename" (...)] redirects all the output of
[(...)] to file "filename.out". This is useful for storing the results of
an [Eval compute], for redirecting the results of a large search, for
automatically generating traces of interesting developments, and so on.
|
|
|
|
equalities in configurations like
x=y
x=z
===
P(x,y,z)
where it now produces
===
P(z,z,z)
In particular (equations are processed from most ancient to most recent).
Thanks to this, a "repeat subst" can just be a "subst" in List.v.
Incidentally: moved a nf_enter to enter in subst_one, since the latter
is normally called from other tactics having normalized evars.
|
|
- add a nra tactic (similar to nia) for non-linear real arithmetic tactic
- fix a long-standing bug in the reification code
- port to the new proof-engine
|
|
- Arithmetic operators and brackets are no longer recognized as bullets,
unless they follow a stop or start a line.
- Most vernacular commands are no longer highlighted when used inside
proof scripts.
- Coqdoc comments now take precedence over regular comments.
|
|
Option -g has no impact on the code generated by GCC, so it is now
systematically added, since it is quite helpful when the VM segfaults
(e.g. bug #4203). Note that, even for a debug build, option -O1 is
preferred to -O0, since -O0 produces assembly code that is much too noisy
for debugging.
For non-debugging builds, -O2 was chosen rather than -O3, since it led to
a noticeably faster VM. I guess even recent GCC compilers still have a
hard time optimizing humongous functions such as the VM.
Here are the results on a simple benchmark: computing the factorial of a
large number with Z and BigZ. (Factorial of 2*6! for Z, of 7! for BigZ.)
For comparison purpose, the timings of native_compute are also provided.
Z BigZ
-O0 6.4 12.3
-O1 4.3 10.7
-O2 2.8 7.3
-O3 3.0 9.3
n_c 2.0 2.4
|
|
|
|
|
|
|
|
Since error messages are ultimately passed to Format, which has its own
buffers for concatenating strings, using concatenation for preparing error
messages just doubles the workload and increases memory pressure.
|
|
|
|
|
|
Followup of: f7b29094fe7cc13ea475447bd30d9a8b942f0fef . In particular, re-closes #3593.
As a side effect, fixes an undiscovered bug of the `eq_constr` tactic which didn't consider terms up to evar instantiation.
|
|
Was made incorrect by 98a710caf5e907344329ee9e9f7b5fd87c50836f .
|
|
Restores the intended behaviour from v8.3 and earlier.
|
|
|
|
|
|
|
|
|
|
as glued.
Possible improvement: rotate using the left children in the glue function,
so that the iter function becomes mostly tail-recursive. Drawback: two
allocations per glue instead of a single one.
This commit makes the following command go from 7.9s to 3.0s:
coqtop <<< "Require Import BigZ ZBinary Reals OrdersEx. Time SearchPattern _." | tail -n 1
|
|
|
|
|
|
This is actually not so perfect because of the lambdas in the return
clause which we would not like to look in.
|
|
for being able to interpret a "match" as a constr pattern).
|
|
in the presence of let-ins).
|
|
|
|
|
|
|
|
them (e.g. "fun ... ⇒ ...") factor well (see #2268).
|
|
We just inline the state in the iolist: less closures makes the GC happier.
|
|
|
|
libraries at once (see #4193).
|
|
|
|
This patch should get rid of the following warning:
Invalid character '-' in identifier "v8-syntax".
|
|
|
|
|
|
The solution is a bit ugly. In order for two tactic notations identifiers not
to be confused by the inclusion from two distinct modules, we embed the name of
the source module in the identifiers. This may still fail if the same module is
included twice with two distinct parameters, but this should not be possible
thanks to the fact any definition in there will forbid the inclusion, for it
would be repeated. People including twice the same empty module otherwise
probably deserve whatever will arise from it.
|
|
Long story short: Filname.concat and other OCaml provided functions
to be "cross platform" don't work for us for two reasons:
1. their behavior is fixed when one compiles ocaml, not when one runs it
2. the build system of Coq is unix only
What is wrong with 1 is that if one compiles ocaml for windows
without requiring cygwin.dll (a good thing, since you don't need to
have cygwin to run ocaml) then the runtime always uses \
as dir_sep, no matter if you are running ocaml from a cygwin shell.
If you use \ as a dir separaton in cygwin command lines, without
going trough the cygpath "mangler", then things go wrong.
The second point is that the makefiles we have need a unix like
environment (e.g. we call sed). So you can't compile Coq without
cygwin, unless you use a different build system, that is not what
we support (1 build system is enough work already, IMO).
To sum up: Running coq/ocaml requires no cygwin, comipling coq requires
a unix like shell, hence paths shall be unix like in configure/build stuff.
|
|
|
|
|
|
|
|
|
|
|
