- 11 2月, 2015 24 次提交
-
-
由 Paul Harrington 提交于
Review feedback from Jason
-
由 Paul Harrington 提交于
-
由 Jared Parsons 提交于
Program for running tests in parallel
-
由 Jared Parsons 提交于
These suites are failing on a number of machines without this change. Disabling.
-
由 Jared Parsons 提交于
-
由 Heejae Chang 提交于
fix test break I introduced.
-
由 Ravi Chande 提交于
Don't preselect enums on backspace
-
由 Ravi Chande 提交于
When backspacing in VB, we used to prefer preselected items over text matches. Dev12 does this for object creation preseletion, but not enum preselection. We should match that behavior. Fixes #287.
-
由 Heejae Chang 提交于
I added encoding check in code that are shared by multiple tests. this change make encoding check to be explicitly requested
-
由 RoslynTeam 提交于
-
由 RoslynTeam 提交于
-
由 RoslynTeam 提交于
Small updates to the change - Sort the output by ellapsed time - Disable a suite which was failing
-
由 Jared Parsons 提交于
-
由 Neal Gafter 提交于
Allow 'is' for static classes for compatibility. Fixes #324.
-
由 RoslynTeam 提交于
Currently tests are run on the developers machine using > msbuild BuildAndTest.proj This takes ~25 minutes to complete on the CI machine and ~45 minutes on a developer machine (developer machines run a greater number of suites). This is not conductive to quick bug fixes. Ideally the infrastructure would switch over to using xunit 2 which can handle parallelizing the tests for us. Unfortunately this isn't an option because: 1. The 32 bit runner quickly OOMS on our suites 2. A number of our tests can't run in 64 bit Long term xunit2 is still our goal. Short term though this tool will use process level parallelization to run our suites.
-
由 Tom Meschter 提交于
Update the DeclarePublicAPI analyzer to consider return types
-
由 Andrew Casey 提交于
Manage the lifetime of DkmClrValue objects
-
由 Matt Warren 提交于
Move VB end construct code fix away from CodeGenerator
-
由 Paul Harrington 提交于
We have two targets that each modify a compiled binary in-place: ApplyOptimizations and FakeSign. They each write out their own sentinel file to help keep incremental build healthy and quick. However, this causes problems when you need to run both (e.g. on a developer box compiling Release binaries). This change moves the sentinel file into a third task that depends on the other two. I hope it's obvious how to extend this if more "post compile binary modification" targets are added later.
-
-
由 Andrew Casey 提交于
We need to attach the same information to DkmEvaluationResults and DkmEvaluationResultEnumContexts, but they have different lifetimes. Enum contexts, which are, in some sense, the children of evaluation results are closed first. As such we don't want to clean up our attached information until the evaluation result is closed. When we are informed that this has happened (by a call to EvalResultDataItem.OnClose), we call Close on the associated DkmClrValue. We hope that actively managing the lifetime of DkmClrValue instances will improve perf by reducing the amount of work that the debugger has to do on each evaluation (which is proportional to the number of live DkmClrValues). Unfortunately, this will not improve perf in the large-array case because DkmEvaluationResults are not closed as they are scrolled out of view. Associated with DevDiv #1094428 (not yet on GitHub).
-
由 Tom Meschter 提交于
The DeclarePublicAPI draws attention to changes to the public API surface area in order to prevent breaking changes and to encourage careful design when adding new APIs. It requires that every project with a public API capture the public types and members in a file called PublicAPI.txt, and creates errors when this file does not match the actual symbols in the project. This makes changes very obvious in code reviews and source control history. Currently PublicAPI.txt does not capture the return types of members (field, properties, methods, and events) so these can changes without it being caught by the tool. This changes updates the analyzer to include the return type. A nice feature of the associated code fix is that it sorts PublicAPI.txt alphabetically, and since type names are fully qualified this naturally groups together types in the same namespace, members in the same type, overloads, etc. To preserve this property we put the return type at the end of the line, after the type and member name.
-
由 Heejae Chang 提交于
while creating pull request, I left out this change
-
由 Neal Gafter 提交于
but continue to report the diagnostic if "strict" mode is requested. Fixes #324
-
- 10 2月, 2015 16 次提交
-
-
由 Heejae Chang 提交于
-
由 Tanner Gooding 提交于
fix vsct file to not have dupliate button definitions
-
由 Tomas Matousek 提交于
Add MetadataVisualizer project to Roslyn solutions
-
由 Andrew Casey 提交于
Construct EnumContextDataItems on demand. Also, don't access their properties before checking for null.
-
由 Andrew Casey 提交于
-
由 Heejae Chang 提交于
disable some tests due to interface tests.
-
由 Heejae Chang 提交于
these tests will be re-enabled once we patch build machine with new interfaces.
-
由 Tomas Matousek 提交于
Conflicts: src/Roslyn.sln
-
由 VSadov 提交于
Replace nonbreaking space (0xA0) with a regular space.
-
由 Andrew Casey 提交于
We need to attach the same information to DkmEvaluationResults and DkmEvaluationResultEnumContexts, but they have different lifetimes. Enum contexts, which are, in some sense, the children of evaluation results are closed first. As such we don't want to clean up our attached information until the evaluation result is closed. When we are informed that this has happened (by a call to EvalResultDataItem.OnClose), we call Close on the associated DkmClrValue. We hope that actively managing the lifetime of DkmClrValue instances will improve perf by reducing the amount of work that the debugger has to do on each evaluation (which is proportional to the number of live DkmClrValues). Unfortunately, this will not improve perf in the large-array case because DkmEvaluationResults are not closed as they are scrolled out of view. Associated with DevDiv #1094428 (not yet on GitHub).
-
由 dino 提交于
-
由 VSadov 提交于
It is likely the character was inserted by accident. There is no reason to use it in the code. It is a valid Unicode whitespace, but not a valid ASCII. Since the containing file has no encoding bytemarks, the behavior of this char is dependent on default encoding. Indeed, the file compiles on Windows and fails on Linux. As far as I could find, this is the only case where this character is used in actual code (not in tests) in Roslyn codebase.
-
由 Matt Warren 提交于
-
-
由 Tomas Matousek 提交于
-
由 Brett Forsgren 提交于
Remove cancellation token from NavigateTo
-