Changing constraint from "struct" to "unmanaged" causes stack overflow and VS crash. #31439
Comments
Korporal
changed the title
|
Do you have a simple repro application that we can use to validate the issue locally? |
|
@tannergooding - That's a pretty fair question, the code however is proprietary and I'm unable to share it. I will however endeavor to create a simple repro, I'm sure I can incrementally copy bits n piece from the original until it happens. The "good" thing here is that it's not intermittent, very consistent. I'm thinking the stack overflow must be a clue to what's happening - by the way, are there any special diagnostic flags I can add to the build or anything that could generate a log for you guys? The "pattern" (if I can say that) is a generic |
|
@Korporal You can use the Windows Event Viewer to get the callstack of the crash |
|
@jmarolf - Ahh OK cool, if I get some time today I'll definitely dig this out and a possible simple repro... |
|
@Korporal event past events should still be there so if you still have the machine it happened on you can get the original crash info [jcouv deleted image as it confused multiple folks ;-) |
|
Seems pretty easy to repro. The following hangs sharplab and crashes VS https://sharplab.io/#v2:EYLgHgbALAPgAgBgARwIwG4CwAoHcDMSAzgC4BOArgMYlIAaAPACoB8OSHSA7gBYCmZPkiZIQSCgDsAtgEMJMgOZ8AJjgDeOAL448hUpRpIAWuvacADmQCWANxkkhjIyyQAzK3wA2yrNm3YgA=== Using using System;
public struct X<T>
where T : unmanaged
{
}
public struct Z
{
private X<Z> field;
} |
|
@tannergooding - Attaboy, you got it - thanks! |
vatsalyaagrawal
added
Bug
Area-IDE
Tenet-Reliability
|
@jcouv Is this related to the signature help changes we worked on recently? |
|
Hi, pleased to see this is readily addressed, very good. May I ask approx when would an update containing this fix expect to be released to the public? Or if this may be several months, is there a workaround by any chance? |
|
I'll first add a test to confirm this is fixed (stack trace looks like previous issue, but still good to validate with a test). I'll do that today. |
|
Correction. This is not fixed and it is not related to the signature help work I had done recently. The issue at hand is a compiler stack overflow with @jaredpar Do you know how to query Watson to see the hit count on this? [Fact, WorkItem(31439, "https://github.com/dotnet/roslyn/issues/31439")]
public void Test()
{
string source = @"
using System;
public struct X<T>
where T : unmanaged
{
}
public struct Z
{
private X<Z> field;
}";
var comp = CreateCompilation(source);
comp.VerifyDiagnostics(); // stack overflow
}Here's the repeating portion of the stacktrace: |
If we can get a bucket ID of the crash it should be possible to query this. |
|
/cc @ivanbasov for help with reliability data |
|
|
@Korporal If all goes well, VS 2019 preview2 (no dates announced). |
|
@jcouv - Hi and thanks. So (just curious) why can't this simply be an update to the compiler and decouple it from a Visual Studio release? I frequntly install updates as we all do and I was under the impression that the compiler is no longer tied to VS releases, indeed we can select the compiler version easily for any project. This is also a rather fundamentql bug, complete failure to process valid C# source. Since VS 2019 appears to expected in mid 2019 this implies the bug won't have a production ready fix for us to use for another 6 months. |
|
In a project, you don't select a "compiler version", you only select a "language version". There is only one version of the compiler installed with VS. That said, Roslyn and the compiler are indeed well componentized inside VS, and it is actually possible to install Roslyn in VS as an extension. This deployment/adoption/dogfooding issue is tracked by #18783 |
|
@jcouv Thanks for explaining this but I must admit I'm still confused! Is this bug special in some way? in the sense that a fix cannot be made and released before VS 2019? I can see many bugs (most being quite obscure) listed here but some of these will presumably be fixed in VS 2017, does my question make sense? Once again, thanks, I'm just trying to understand the situation. |
|
Every bug is triaged and prioritized based on impact (how common is it, how bad is the experience), risk of change, and some other factors (regression, compatibility, etc). The bar for fixing issues in VS 2017 is very high at this point, and this bug does not meet that bar. |
|
Thanks, that explains things, much appreciated. |
|
Moved back to preview3. Check with Jared and move out again if ok. Thanks |
* Add regression test for issue #31439 * Delay constraint checking on field types until after type members are added * Fix tests * Fix WithNullability for field type constraint checks * Use SyntaxTree property instead of TypeSyntax.SyntaxTree * Fix up more expected diagnostics
|
@RikkiGibson - Hi, is this fixed in the C# 8.0 beta that's part of VS 2019 16.0.0 preview 4.4? Attempting to compile this generates a compilation error (the crash has gone but it seems there are still issues): namespace ConsoleApp3
{
class Program
{
static void Main(string[] args)
{
}
}
public struct X<T> where T : unmanaged
{
}
public struct Z
{
private X<Z> field;
}
} |
|
@jmarolf - Hi, I'm not familair with sharplab, looks pretty useful but each time I visit I see a strange error messages at the top of the page:
and I have no idea what that means, the left hand side edit window becomes greyed out too. |
|
It can be finicky occasionally.. I suggest just refreshing, maybe clearing caches and stuff, and report it in https://github.com/ashmind/sharplab if it keeps happening. |
|
@RikkiGibson @jmarolf @jcouv - Hi. There's is considerable confusion here - I cannot get an answer from anyone in the csharplang repo as to whether the code fragment should actually compile (is legal C#) or if it should fail to compile and if it should fail then the error message:
is an appropriate error message. See this issue which is a recent history of the discussion with the csharplang techies. Is there anyone who knows enough about the language and its formal specification who can give a definitive yes/no answer? My own understanding is that the code should compile when the Surely this is a simple matter of examining the formal language specification? after all a fix was made by @RikkiGibson (as stated in this issue thread) so that fix must have been to make the compiler align with said specification, yes? Thanks! |
|
Apologies for the delay. I suspect the hang you're experiencing in SharpLab is due to using the 2.9.0 compiler instead of the master compiler (select 'master' from the dropdown list). Sometimes when the compiler stack overflows on the server side, SharpLab will just hang. Now, regarding the sample: public struct X<T> where T : unmanaged
{
}
public struct Z
{
private X<Z> field; // error in 7.3: feature 'unmanaged constructed types' requires C# 8.0
}This is not valid in C# 7.3 and earlier. The reason is that Z is considered 'managed' by that language version due to containing a field of a constructed type (namely However, this is valid in C# 8. The unmanaged constructed types feature (see dotnet/csharplang#1744) relaxes the rules to allow types such as |
|
It is valid in the current master bits and can be tested by pulling down a nightly build of the compiler. For example: https://sharplab.io/#v2:EYLgtghgzgLgpgJwD4AEDMACWCCuBjGDADQB4AVAPgwHcALRODMjEDHAO0nYgHM4ATALAAoAN4iAviJHosMXAQwAtEeOEYNGAA4IAlgDcI8YiSVUAZrrgAbfgG4MAekcZECAPYIMu9hgDsAHRorOZwRjgIjADkHFy8Ahh47uzY+PD8GDAAnlpwUFEYkQCOOLqRUBgAwgDEGAAcAQAMkkA=== C# 8 is not RTM and the compiler that shipped in VS2019 only provides C# 8 in a preview form ( |
…et#32536) * Add regression test for issue dotnet#31439 * Delay constraint checking on field types until after type members are added * Fix tests * Fix WithNullability for field type constraint checks * Use SyntaxTree property instead of TypeSyntax.SyntaxTree * Fix up more expected diagnostics
I'm working on some code in VS 2017 (v 15.9.2) using v 7.3 of C#. This is a small console app but does some low level memory manipulation. I was refactoring the code earlier and decided that a constraint could be adjusted to
unmanagedrather thanstructto support some richer operations.Before this change the code compiles and runs exactly as expected. The code here is basically a generic
structwhere theTis itself constrained to bestruct- changing this constraint tounmanagedis what leads to the failure.I then decided to edit
Program.csin notepad just to change the constraint and then re-opened the solution in VS. I opened the solution (but notProgram.cs) and ran a build, the build soon failed with stack overflow.I then opened
Program.csagain (in VS 2017) and VS simply freezes, stops interacting and responding to mouse etc. Left for long enough it abruptly dies and restarts.I did explore the
unmanagedconstraint recently but in a simple test app and probably used an earlier version of C# than 7.3 - I had no issues. So I simply opened the solution and changed each project to C# 7.2 but again as soon as I opened Program.cs VS freezes, same thing happens when I try C# 7.1 and C# 7.0 too.NOTE: This is not a runtime issue, so don't suspect that the logic itself is involved, this is an IDE/compiler issue.
Thoughts?
The text was updated successfully, but these errors were encountered: