- 23 1月, 2020 4 次提交
-
-
由 Ivan Basov 提交于
-
由 Allison Chou 提交于
-
由 Ivan Basov 提交于
-
由 Allison Chou 提交于
-
- 22 1月, 2020 10 次提交
-
-
由 Allison Chou 提交于
-
由 Ivan Basov 提交于
-
由 Ivan Basov 提交于
-
由 Ivan Basov 提交于
-
由 Ivan Basov 提交于
-
由 Ivan Basov 提交于
-
由 Ivan Basov 提交于
-
由 David Poeschl 提交于
-
由 Ivan Basov 提交于
-
由 Allison Chou 提交于
-
- 21 1月, 2020 2 次提交
-
-
由 Tomáš Matoušek 提交于
-
由 Tomáš Matoušek 提交于
Previously we reported diagnostics caused by the debuggee not allowing edits in the corresponding module in EnC diagnostic analyzer. A module does not support edits if any of its instances does not support the edits. There are scenarios in which module instances may be loaded while in break mode. E.g. if the user attaches to a secondary process while debugging a primary process, the secondary process might have a module instance loaded of the same module loaded to the primary process. To handle these cases we hooked up debugger module load/unload events and invalidated the reported diagnostics. This approach has a few problems: 1) EnC analyzer is now non-deterministic, as it can give different answers for the same inputs depending on the state of the debuggee 2) Turns out that the debugger does not deliver module load/unload events in all cases (e.g. when the secondary process is killed). 3) It's a bit complicated and would get even more complicated once we moved EnC to OOP/Cloud Instead, we don't report these errors until the user triggers the "continue" action. At this point the VS and all processes being debugged are blocked until the changes analyzed and applied. Since this situation is not very common the slight regression in interactivity is acceptable.
-
- 19 1月, 2020 7 次提交
-
-
由 Siegfried Pammer 提交于
-
由 Cyrus Najmabadi 提交于
-
由 Cyrus Najmabadi 提交于
-
由 Cyrus Najmabadi 提交于
-
由 Cyrus Najmabadi 提交于
These identifiers are already simplified as part of nodes further up the tree.
-
由 Cyrus Najmabadi 提交于
-
由 CyrusNajmabadi 提交于
-
- 18 1月, 2020 17 次提交
-
-
由 Siegfried Pammer 提交于
-
由 Allison Chou 提交于
-
由 Julien Couvreur 提交于
-
由 Ivan Basov 提交于
-
由 Ivan Basov 提交于
-
由 Allison Chou 提交于
-
由 Allison Chou 提交于
-
由 Cyrus Najmabadi 提交于
-
由 Cyrus Najmabadi 提交于
-
由 Andy Gocke 提交于
Capturing lambdas should be avoided in core compiler code as they generate excess allocations Fixes #41054
-
由 Fred Silberberg 提交于
* Insert correct `WithNullableContext` binder into the binder chain to ensure speculative models return correct information based on speuclation location. * Refactor Nullable context options to rely on an enum instead of two bools. Break out the BuckStopsHereBinder's code to allow sharing. Fix binder ordering issue for TryGetSpeculativeSemanticModelForMethodBody. * Make NullableContextState.State a byte for space.
-
由 Sam Harwell 提交于
Addresses 3200MB allocations while analyzing Roslyn.sln.
-
由 Sam Harwell 提交于
Addresses 950MB allocations while analyzing Roslyn.sln.
-
由 Sam Harwell 提交于
Addresses 950MB allocations while analyzing Roslyn.sln.
-
由 Sam Harwell 提交于
Addresses 2400MB allocations while analyzing Roslyn.sln.
-
由 David Poeschl 提交于
-
由 David Poeschl 提交于
-