Compilation Takes Significant Amount of Time To Finish

[JasonBock]

Version Used:
.NET 10

FWIW, on my machine:

dotnet --list-sdks
9.0.311 [C:\Program Files\dotnet\sdk]
10.0.103 [C:\Program Files\dotnet\sdk]
10.0.200-preview.0.26103.119 [C:\Program Files\dotnet\sdk]

Steps to Reproduce:

1. Go to this branch of the repository and clone it: https://github.com/JasonBock/NumberTheoreticTransform/tree/1-Create-Working-Version-of-Original-Code
2. Build the solution: dotnet build .\NumberTheoreticTransform.slnx

Expected Behavior:
The code builds in seconds.

Actual Behavior:
The code builds in 16 minutes:

To give a bit of background on this code, this is basically the code I wrote for my master's thesis in 1995. That was done in C, and what I'm trying to do is to reproduce it as closely as I can in C#, mostly for historical purposes. The code is bad - like, really bad - as I really didn't know what I was doing 30 years ago :). But, I was hoping I could do a decent 1-to-1-ish translation. The end result is a Program.cs with about 2.4K LOCs, but I was kind of shocked to see just how long it would take to build this. (I know the C code in 1995 didn't take this long to compile!).

The computer I'm building this on...without getting into specs, it's only a year or so old, and all of the C# development I've done on it has always compiled in the "seconds" range. Never seen anything take this long.

I'm guessing there's something to the structure of the code that's doing ... something that Roslyn doesn't really like. The mpmult() function has a bunch of nested for loops, maybe that's the issue. I don't think this is necessarily uncovering a bug in Roslyn, but it is a curiously that the compile time is as long as it is. If there's something I need to restructure to make the build go faster, I'm willing to do that....just not sure where to start with this.

(Side note: The resulting EXE has an exception when it runs, but that's something I'm 99.999% confident is in my code that I need to fix, and not an issue with the compiler itself. However, given how long it takes to compile, I haven't spent time trying to figure out what the issue is.)

[RikkiGibson]

I'm curious how it performs if you set <Nullable>disable</Nullable>, it sometimes has issues with huge methods.

[JasonBock]

@RikkiGibson Tried that, it dropped it to 539.1 seconds, but that's still 9 minutes, which is a lot longer than I would expect.

[jhinder]

It appears to be related to analyzers, not the compiler itself. If you pass /p:RunAnalyzers=false (or set the equivalent property), the build time drops down into the second range (the most I've seen in my tests was 2.3s):

D:\repos\NumberTheoreticTransform\src>dotnet build /p:RunAnalyzers=false NumberTheoreticTransform.slnx
Restore complete (0,4s)
    info NETSDK1057: You are using a preview version of .NET. See: https://aka.ms/dotnet-support-policy
  NumberTheoreticTransform net10.0 succeeded (0,1s) → artifacts\bin\NumberTheoreticTransform\debug\NumberTheoreticTransform.dll

Build succeeded in 0,6s

I've run a build with analyzer reporting enabled (/p:ReportAnalyzer=true), and one stands out above the rest by a large margin: Microsoft.CodeQuality.CSharp.Analyzers.Maintainability.CSharpAvoidDeadConditionalCode (CA1508).

Screenshot from the MSBuild Log Viewer:

[JasonBock]

@jhinder I was wondering last night before I feel asleep if it was due to an analyzer, or group of analyzers. By default I always set <TreatWarningsAsErrors>True</TreatWarningsAsErrors>, but I'm fine with suppressing analyzers if they don't make sense. I added <NoWarn>$(NoWarn);CA1508</NoWarn> to my project, and now the build takes 1 second.

I'll leave it to the powers that be here if someone should look at CA1508 to see if there's something in its implementation that can be improved.

[JasonBock]

Looks like there's at least one open issue on CA1508 - dotnet/roslyn-analyzers#7125