Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Backports for 1.6-beta #38795

Merged
merged 57 commits into from
Dec 19, 2020
Merged

Backports for 1.6-beta #38795

merged 57 commits into from
Dec 19, 2020

Conversation

KristofferC
Copy link
Member

@KristofferC KristofferC commented Dec 9, 2020

Backported PRs:

Non-merged PRs with backport label:

Fixes #38758

(cherry picked from commit 5f1b21b)
@KristofferC KristofferC added the release Release management and versioning. label Dec 9, 2020
DilumAluthgeBot and others added 12 commits December 11, 2020 10:21
Co-authored-by: Dilum Aluthge <dilum@aluthge.com>
(cherry picked from commit e4da832)
…8796)

We switched away from using MbedTLS on all platforms to use the native
TLS libraries on Windows/MacOS; let's ensure that a from-source build
does the same here as well.

(cherry picked from commit b407c5a)
…38809)

Co-authored-by: Dilum Aluthge <dilum@aluthge.com>
(cherry picked from commit 36fbb9a)
Co-authored-by: Dilum Aluthge <dilum@aluthge.com>
(cherry picked from commit edfcd5a)
The initial draft of Fake JLLs missed some important points in JLL
API compatibility; let's close the gap somewhat, where we can.  This PR
adds the following exports:

* `get_artifact_dir()`: returns the Julia prefix
* `dev_jll()`: Throws an error
* `best_wrapper`: always set to `nothing`
* `get_*_path()`: returns the path of the identified library product

It also converts all fake JLL modules to `baremodule`s, and sets the
appropriate compiler options to minimize compilation.

(cherry picked from commit d8975fa)
Also, remove DocumenterLaTeX, since the PDF/LaTeX backend is now again
provided by Documenter directly.

(cherry picked from commit b7d5240)
This doesn't do much currently, because we only call this function
on `Const` objects which we don't currently create if the initialization
of the object is incomplete, but we may want to do so in the future,
so might as well be defensive about it.

(cherry picked from commit 1327b5c)
Since we can inline pointer-containing structs into other structs now,
an `isptr` check is insufficient to determine whether or not we need
to recurse here. Also check the actual type of the field in addition.

(cherry picked from commit 793f875)
Co-authored-by: Daniel Karrasch <daniel.karrasch@posteo.de>
(cherry picked from commit b1a2847)
Introduced in #38389

(cherry picked from commit d6f9948)
vchuravy and others added 4 commits December 12, 2020 13:19
@KristofferC KristofferC changed the base branch from master to release-1.6 December 12, 2020 18:23
vchuravy and others added 10 commits December 12, 2020 15:56
This avoids a dirty repo state after building on macOS.

(cherry picked from commit e10d7a3)
* improve inferrability within TOML module

* simplify with `@try` macro

* apply suggestion, use `Int64`

(cherry picked from commit b3eaa34)
* Adjust calling convetion of dgebal and co

* Adjust calling convention of BLAS & LAPACK

(cherry picked from commit 9f4a807)
Co-authored-by: EternalLearner42 <46832510+EternalLearner42@users.noreply.github.com>
(cherry picked from commit d34cc20)
vchuravy and others added 26 commits December 17, 2020 10:21
This also fixes an insecure behavior: even if `set_ssl_cert_locations`
failed, `REFCOUNT` was still incremented, which meant that subsequent
calls to `ensure_initialized` didn't call `initialize` and so there was
never a successful call to `set_ssl_cert_locations`. Without this
libgit2 defaults to not verifying host identities and that is insecure.
To prevent this, this patch locks on `ensure_initialized` and decrements
`REFCOUNT` if initialize throws an error, ensuring that `initialize`
succeeds at least once, including the call to `set_ssl_cert_locations`.

(cherry picked from commit 4dede6d)
Also change "equivalent to" to a more rough "akin to" since `x .op y` isn't always identical to `broadcast(op, x, y)`.

(cherry picked from commit 74a08fe)
(cherry picked from commit 2362037)
* Add section regarding Pkg mode to REPL docs

Co-authored-by: Stephen Vavasis <vavasis@uwaterloo.ca>
(cherry picked from commit ce338db)
pcre2-cet-flags.patch lacks some trailing whitespace that exists in
one of the target files. Consequently, the patch does not apply when
the patch tool is strict about whitespace.

This teensy diff makes the whitespace in pcre2-cet-flags.patch
strictly match the whitespace in the target file.

Co-Authored-By: Nathan Daly <NHDaly@gmail.com>

Co-authored-by: Nathan Daly <NHDaly@gmail.com>
(cherry picked from commit 3853060)
Probably needs WTF-8 support to be correct, but may avoid crashing.

Fix #38838

(cherry picked from commit ec17c69)
(cherry picked from commit a6f0e69)
On some platforms (PowerPC) the call to the `jlplt` is not a tail-call
and so it will be part of the backtrace. This means we are off-by-one
and won't skip the Julia function `backtrace` messing up tests that
check precise formatting.

(cherry picked from commit fc577d0)
* Update non-BB dsfmt build to match with the BB one.
Update URLs to https
Fix #17945

* Add -DDSFMT_SHLIB

(cherry picked from commit b36338a)
Perhaps the most annoying thing about Revise now is that the first
revision is quite slow, about 3.1s on my machine. This PR drops the time
to about 2.4s. Basically the idea is to precompile statements that
Revise will need.

Discovered via the new snoopi_deep/Core.Compiler.Timings framework.

(cherry picked from commit 4c4e078)
For some reason (perhaps #32705?) most or all of these fail if they
are emitted as precompile statements, so this moves them into Base itself.

This drops the time for a revision down to 1.85s.

(cherry picked from commit cc1623b)
…#38918)

* Document the need for disabling BB if building without network access

Fix #33569

* Update doc/build/build.md

Co-authored-by: Dilum Aluthge <dilum@aluthge.com>

Co-authored-by: Keno Fischer <keno@juliacomputing.com>
Co-authored-by: Dilum Aluthge <dilum@aluthge.com>
(cherry picked from commit 9c2d813)
* Improve position of MKL.jl mention

* Clarify that SuiteSparse will not use MKL if going down the MKL.jl path

Co-authored-by: Viral B. Shah <ViralBShah@users.noreply.github.com>
(cherry picked from commit 996dd55)
This was a copy-paste error.

(cherry picked from commit a2f0337)
…ructions on Windows (#38910)

* Delete winrpm, vagrant and appveyor build files

* Update Windows build instructions

(cherry picked from commit 9e27eee)
(cherry picked from commit 8a84395)
@KristofferC
Copy link
Member Author

The commit list is getting a bit long here and it gets messy, so I'll just merge this, remove the label from the backported PRs and continue in another PR.

@KristofferC KristofferC changed the title WIP: Backports for 1.6-beta Backports for 1.6-beta Dec 19, 2020
@KristofferC KristofferC merged commit de69b02 into release-1.6 Dec 19, 2020
@KristofferC KristofferC deleted the backports-release-1.6 branch December 19, 2020 19:43
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
release Release management and versioning.
Projects
None yet
Development

Successfully merging this pull request may close these issues.

Stack overflow when defining some promote_rules on nightly