GitLab

* legacy: output --preprocess result on ExecStdout

* legacy: removed redundant argument filters

* legacy: moving path-relativization code out of PrepareCommand

* legacy: removed i18n on unneded message

* legacy: replacing ParseCommandLine with the equivalent library call

* legacy: removed parameter that happens to be always false

* legacy: removed constants.MSG_PATTERN_MISSING from i18n

* Let ExecRecipe return the command executed

This prepares for building a compilation database later. The returned
command is not currently used anywhere yet, so this commit should not
change behaviour.

* Fixed typo
Co-authored-by: Matthijs Kooijman <matthijs@stdin.nl>

builder: use ar-chives for linking big sketches (was: use the @ syntax to reduce command line length if needed) (#961)

* Small cosmetic changes

* Create an archive file if the number of object files is too big

This should fix the command line too big issue on Windows.

* fixed comments

* When exploiting ar-chives make a .a file for each soruce subfolder

This is required because gcc-ar checks if an object file is already
in the archive by looking ONLY at the filename WITHOUT the path, so
it may happens that, for example, an object file named "subdir/spi.o",
already inside the archive, may be overwritten by an object file in
"anotherdir/spi.o" only because they are both named spi.o and even if
they are compiled on different directories.

* using paths.PathList to keep objectFileList

* Add support for definitions containing quotes in compile --build-properties flag

* Deprecate --build-properties flag in favor of --build-property

* Fix some escaping issues on Windows

* Simplify compile examples

* Add zip library imports
* Add git library imports
* Add tests for lib install with flags

Running command through "go run ..." will introduce artificial delayes.

* Added 'debug --info/-I' flag to get debug metadata

* Debugger: do not use anymore pre-made recipes

* Added nicer color ouput for 'debug --info'

* Removed deprecated field

* Merged together DebugConfigReq protoc type

It makes no sense to have two separate types to create a debug session
or to create a debug configuration.

* Debug: added programmer selection flag

* debug: return error if debugging is not supported by the platform

* debug: openocd server now require full path to openocd executable

* debug: fixed test runner

* debugger: Fixed command line tests

* debugger: hotfix for samd platforms 1.8.8 and 1.8.9

The `core update-index` command was incorrectly written as "core update index".

* Added UploadUsingProgrammer gRPC call

* fix indent

* Added more tests for upload

* Added test for upload without port

* Added test for upload without programmer (with and w/o port)

* Added test for burn bootloader

* Added test for burn bootloader with bootloader.tool options override

* Increased verbosity

* Refactored upload procedure to correctly handle bootloader.tool property

* Handle windows quirks on command output

* Simplified tests

[skip-changelog] Added `discovery.Discovery` object to handle communication with pluggable discoveries (#1029)

* Pluggable Discovery handler: first implementation

* Added 'discovery_client' (for debugging discoveries)

* Added 'event' mode in PluggableDiscovery

* discovery_client now supports multiple discoveries

* Added ID to PluggableDiscovery

* Added PluggableDiscovery.String() implementation

* Fixed TestDiscoveryStdioHandling

* Fixed discovery test run on Windows

It really takes that long for messages to go back and forth.
I don't know if there is a simpler way to reduce stdio buffering time.

Previously, release branch creation did not trigger documentation publishing.
Co-authored-by: per1234 <accounts@perglass.com>

* Sligthly simplified upload test cases handling

* Make --input-dir handle gracefully some rare cases

Using `--input-dir path/to/firmware` with a directory containing:

  path/to/firmware/firmware.ino.bin
  path/to/firmware/some_other_firmware_name.ino.bin

should not fail but select `firmware.ino.bin` for upload because the
containing folder has the same name.

See https://github.com/arduino/arduino-cli/issues/765#issuecomment-699678646

Co-authored-by: Cristian Maglie <c.maglie@arduino.cc>

* Fix platform tool dependency determination

The platform tool dependency check was giving false negatives due to comparing pointers instead of version values.

This caused tools to be removed during platform uninstallation even when another installed platform had a dependency on
that tool.

* Add integration test for tool removal during platform uninstall

Tool dependency by `arduino-cli core uninstall` is somewhat complex because it must only uninstall the tools that no
other platforms have a dependency on. If it doesn't, it breaks the other platform and the cause of this breakage would
likely not be obvious to the user.

So it's important to test to ensure this functionality continues to work correctly.

This reverts commit 900654f4.

Flake8/pycodestyle has a pair of mutually exclusive checks:

- W503: 	line break before binary operator
- W504: 	line break after binary operator

Having both these checks enabled results in a failed check when there is a line break at a binary operator that can't be
resolved by moving the operator.

PEP 8 recommends line break before the binary operator, so the logical choice between the two checks is W504.

The "Build process" section of the Arduino platform specification contains a short summary of the Arduino sketch build
process. The "Sketch build process" page of the documentation provides much more detailed information, which wouldn't be
appropriate in the platform specification, but could well be of interest to a platform developer.

Adding a link to the "Sketch build process" documentation from that section of the Platform specification makes all
relevant information easily accessible to the reader.

The old tool definition example in the package_index.json specification was quite outdated. It did not demonstrate the
ARM hosts that have become more important since the specification was written. The macOS host value was antiquated.

This could lead to the reader being left with no clear idea of how to create the tool definitions for those hosts.

upload.maximum_data_size and upload.maximum_size are properties used by the build system to fail
compilation when memory usage exceeds a threshold and to calculate the relative memory usage for display in the console.

Pull requests are only one of the possible ways to contribute to Arduino CLI, yet they were the only one mentioned in
the "how to contribute" documentation. This change adds documentation of other options, providing more opportunities to
assist in this project.

Previously, instructions for running integration tests were present both in CONTRIBUTING.md as well as
in test/README.md.

Although most of the content was duplicated, there was
some important unique content on each page and the readers of one document would not necessarily know
of the existence of relevant information in the other.

Duplicate content is more work to maintain, and likely to end up becoming outdated.

So I have transfered the unique content from test/README.md to CONTRIBUTING.md and replaced it with a
link to that content.