Give a different error to guide use to install via global tools so, if several bundled DotnetTools cannot finish source build on time. The user can use global tools to get it.
The original plan that adding a different resolver is hard due to resolver can only find dll that will be used to spawn a process. However, the command constructor will give an error message when resolver find null. By adding a different error when the command name is part of the list, it can achieve the same goal.
If there are shims packaged by convention in nupkg. Shim Repository will simply copy it to the right location.
The query interface ToolPackageInstance will be in charge of finding the shim folder and filter the right RID. Shim Repository will pick the right file after the folder is located since Shim Repository knows the shim name and it also book keep the files at uninstallation.
During development, due to the wrong adapter level. The mock duplicated too much logic. So, I corrected the abstraction level to lower (only create shim). And replaced the existing mock with a much smaller one without any atomic control and file move, copy logic. At the same time. The chmod, which is a IO action, causes problem during tests. So I added adapter layer to it and put it in Util.
On environments where registry access is disabled, the first run experience
fails because it could not add the tools path to the user's environment.
This fix properly handles the security exception by printing a warning and
continuing. Users will have to manually add the PATH environment variable to
their environments to prevent `dotnet tool install` from printing PATH
instructions.
A new file sentinel is added to track whether or not the PATH has been
modified. The first run experience also now correctly skips modifying the PATH
if `DOTNET_SKIP_FIRST_TIME_EXPERIENCE` is set.
Fixes#8874.
This commit checks that the `--tool-path` option for the `tool list` and `tool
uninstall` commands is a directory that exists.
If the directory does not exist, an error and the command help is displayed.
Fixes#8931.
* Publish app host to folder under SDK
* Use carried apphost as shim
* Remove full framework launcher
* Fix test run command issue
* Use latest release/2.1 build
* Test with 32 bit env
* Add missing return
* Update to latest prodcon build
* Add xlfs
This commit implements the `buildserver shutdown` command that can be used to
shutdown MSBuild, VB/C# compiler, and Razor build servers.
By default, all three build servers are shut down. Options can be passed to
shut down a subset of the build servers.
Fixes#8185.
This commit ensures the correct property (`ProjectTypeGuids`) is respected when
adding a project to a solution file.
Additionally, we now error if a project type GUID cannot be determined rather
than incorrectly mapping to the C# project type.
Enabled previously disabled tests that were waiting on upstream changes from
MSBuild and F#.
Fixes#5131.
Fixes#7742.
This commit implements the missing `--tool-path` option for the list tool
command. This enables the command to list locally installed tools.
Fixes#8803.
This commit fixes the tool package store such that it stores a full path
instead of, potentially, a relative path.
This prevents a relative path from inadvertently being passed to NuGet
during the restore and causing it to restore relative to the temp project
directory.
Fixes#8829.
Currently the list tool command tests, while localizing the column headers,
didn't properly take into account the fact that localized builds might produce
strings longer than the English versions of the column header strings. This
results in a mismatch of the actual from the expected due to additional column
padding.
The fix is to stop using a static expected table and do a simple calculation of
the expected table based on the length of the localized strings.
Fixes issue related to PR #8799.
Other than change source to source-feed and make it additional instead of exclusive. I changed source to be multiple. Because restore support multiple source https://github.com/Microsoft/dotnet/issues/361
As for mock. The offline feed and source feed is considered the same, so remove the category of “source”. I renamed source to “AdditionalFeed” because that is more accurate on implementation level.
Note:
NuGet feed don’t have order. Whichever responses the fastest, is the first.
No change on restore.
scripts/cli-test-env.sh change is due to mac 10.13 is finally added to RID graph. And it is “considered” one of the CLI supported RID
This commit fixes the ToolPackageInstaller tests so that they no longer modify
the current working directory. The directory being set is now being properly
passed in as an argument to override the default of the current working
directory.
Additionally, this commit also changes the package root to a temp location
rather than based off of the current working directory.
This commit attempts to filter the diagnostic messages emitted during tool
installation. The diagnostic messages may be prefixed with the temporary
project; since this is an implementation detail that only causes confusion and
clutter in the diagnostic messages, the prefix is removed if present.
Fixes#8707.
* Change to escape string via XML
* tool-path option -- "Session tool"
From the beginning design, shim and packageInstaller take package location from constructor and don't have assumption anymore. From previous discussion, tool-path will simply change global location to the one user want, and everything else is the same.
However, this "override" need to happen during the call, that means InstallToolCommand will create different shim and packageInstaller object according to the tool-path during the call instead of constructor DI.
* global package location change
* block of leading dot as command name
* Localization of tool-path option
