| Age | Commit message (Collapse) | Author |
|---|
|
Not much benefit and it breaks make's print-directory system.
|
|
They should not be necessary today as they date from the shareable
pre-artifact epoch, an incur in an slowdown.
|
|
Also fixes is_in_projects
|
|
|
|
incidentally the "projects" array can be queried to get the list
of projects
|
|
|
|
Git wants an identity and none is setup on the CI.
|
|
This avoids the need to rebase the overlay when nothing has changed.
|
|
|
|
Also pass `--output-sync` on the CI, as suggested in
https://github.com/coq/coq/pull/12653#issuecomment-696226093, to protect
against this failure mode.
Fixes #13062
|
|
- don't `set -x` while loading overlays, non-verbose untaring
- ls _build_ci to help figure out artifact download issues
|
|
|
|
We were incorrectly calling the global `install` target even when
building only subcomponents of the library.
|
|
This ensures that the log will contain the commit hash that we tested.
This reuses the method from the Windows build script (we have a number
of common functions, it would be interesting to start a bash shared
library file).
|
|
This is very preliminary but you should get the idea. The script tries
to build contribs, then creates a branch and sets the remote properly
as to submit a PR.
Usage example:
```
./dev/tools/create_overlays.sh ejgallego 9999 ltac2 elpi
```
This only works for contributions hosted in github for now.
|
|
This was kept as a fallback for some time, not worth keeping it
anymore as our GitLab setup seems mature and reliable enough.
|
|
This should allow people to test CI contribs under Dune. It should be
good for now but it is still a work in progress.
|
|
|
|
|
|
This allows to use fixed commits and not just branches or tags.
We keep using git clone when $FORCE_GIT is set (for projects on
gforge.inria.fr and projects pulling dependencies through git submodules).
fiat-parsers also calls git submodule, but inside its own Makefile.
|
|
This has become mostly garbage since GitLab CI became our main CI platform.
|
|
GitLab setup is quite stable these days thanks to the work of many
people and `coqbot`. We decided to keep CircleCI support for a while
as a safeguard in case something happened in the migration to GitLab,
but these days we are just wasting resources to them and to us. As I'm
afraid CircleCI won't scale for us, the time to remove it has arrived.
Still, CircleCI had some awesome functionality that GitLab's CI
doesn't offer yet, see the links at:
https://github.com/coq/coq/issues/6919#issuecomment-395885573
- https://gitlab.com/gitlab-org/gitlab-ce/issues/29347
- https://gitlab.com/gitlab-org/gitlab-ce/issues/35222
- https://gitlab.com/gitlab-org/gitlab-ce/issues/41947
- https://gitlab.com/gitlab-org/gitlab-ce/issues/47063
|
|
|
|
|
|
This is needed for CI packages that use `META.coq` such as in
https://github.com/coq/coq/pull/7656 .
|
|
|
|
In the original Travis CI setup, the per-job time limit was an
issue. However, Gitlab has much improved this problem due to
a) Coq not being built for each contrib,
b) user-configurable time limit.
We thus disable the expensive builds from Travis:
`fiat-crypto`, `formal-topology`, `geocoq`, `iris-lambda-rust`,
`math-comp`, `unimath`, `vst`
and instruct Gitlab to build [`geocoq`, `math-comp`, `unimath`, `vst`]
in full.
We also update the `math-comp` script as the `odd-order` theorem lives
in a separate repository and it is a key CI case.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
When invoking through Makefile we always rebuild dependencies. To skip
dependencies, invoke ci-wrapper directly. We make Circle CI do this.
In order to properly support invoking ci-wrapper directly we replace
"make" in ci-common by a bash function which adds -j to the make
invocation outside submakes. We also set TIMED in the ci-wrapper.
|
|
|
|
- timing needs time and python
- check for compiled files without source looks in the install
directory (except for make -f Makefile.ci which doesn't check), as
such the install directory has been renamed to _install_ci and isn't
searched.
|
|
|
|
According to
https://www.gnu.org/software/make/manual/html_node/Options_002fRecursion.html#Options_002fRecursion
it's not necessary, because we pass `-j ${NJOBS}` to the top-level `make`
invocation in `.travis.yml`. Additionally, explicitly passing `-j` in,
e.g., fiat-crypto, results in error messages such as
```
make[2]: *** write jobserver: Bad file descriptor. Stop.
make[2]: *** Waiting for unfinished jobs....
make[2]: *** write jobserver: Bad file descriptor. Stop.
make[1]: *** [coqprime] Error 2
make[1]: INTERNAL: Exiting with 1 jobserver tokens available; should be 2!
make[1]: Leaving directory `/home/travis/build/JasonGross/coq/_build_ci/fiat-c
```
because the `-j` on the `make` in the `ci-fiat-crypto.sh` script
disables jobserver mode, and the submake in fiat-crypto to make coqprime
does not explicitly pass `-j`, and so reenables jobserver mode, and then
`make` gets very confused.
Commit made with
```bash
cd dev/ci
git grep --name-only -- 'make -j ${NJOBS}' | xargs sed s'/make -j \${NJOBS}/make/g' -i
git grep --name-only -- 'make -f Makefile.coq -j ${NJOBS}' | xargs sed s'/make -f Makefile.coq -j \${NJOBS}/make -f Makefile.coq/g' -i
```
|
|
This will avoid stupid merge conflicts in the future.
|
|
|
|
|
|
We need to do a bit of hacking, but it should be fine for the short
term.
c.f. https://gitlab.mpi-sws.org/FP/iris-coq/issues/83
|
|
|
|
We now allow the user to overlay contribution repositories and
branches by adding their own rules to `ci-basic-overlay.sh`.
This just provides very basic support.
|
|
This commit also documents the behavior of said function;
and fix the location of the ssreflect clone to an absolute
path (this is now necessary).
|
|
|
|
- Move the git clones to a specific subfolder to avoid pollution.
- Do not fail when git clone already exist (but make sure it is up-to-date).
|
|
|
|
|
