| 23 Sep 2025 |
| @jolman:matrix.org left the room. | 20:12:34 |
| 24 Sep 2025 |
Whovian9369 | Here to cause trouble again - I'm trying to bump a project from NET 8 to NET 9 in a derivation (without code changes, dev says it should work), but that Nuke build process seems to fail due to using a dynamically linked executable... Hapen to have any immediate ideas?
Could not start dynamically linked executable: /tmp/fetch-deps-hactoolnet-2024.08.29.Pzmhk6/source/build/bin/Debug/_build (Then the message is about stub-ld etc)
| 02:39:09 |
Whovian9369 | * Here to cause trouble again - I'm trying to bump a project from NET 8 to NET 9 in a derivation (without code changes, dev says it should work), but that Nuke build process seems to fail due to using a dynamically linked executable... Hapen to have any immediate ideas?
Could not start dynamically linked executable: /tmp/fetch-deps-hactoolnet-2024.08.29.Pzmhk6/source/build/bin/Debug/_build
(Then the message is about stub-ld etc)
| 02:39:19 |
Whovian9369 | Alternatively, is there a way to keep the in-progress /tmp/fetch-deps-... path even after a failed run? I feel like that would be good for testing. | 02:39:55 |
Whovian9369 | * Alternatively, is there a way to keep the in-progress /tmp/fetch-deps-... path even after a failed run? I feel like that would be good for testing.
The same process works fine on Net 8 but not Net 9, so I'm stumped from what I can tell.
preConfigure = ''
dotnet --version > DotnetCliVersion.txt
patchShebangs --build build.sh
for rid in $dotnetRuntimeIds; do dotnet restore --runtime "$rid" "build/_build.csproj"; done
./build.sh Codegen --configuration ${buildType}
'';
| 02:42:47 |
Whovian9369 | * Alternatively, is there a way to keep the in-progress /tmp/fetch-deps-... path even after a failed run? I feel like that would be good for testing.
The same process works fine on Net 8 but not Net 9, so I'm stumped from what I can tell.
{
preConfigure = ''
dotnet --version > DotnetCliVersion.txt
patchShebangs --build build.sh
for rid in $dotnetRuntimeIds; do dotnet restore --runtime "$rid" "build/_build.csproj"; done
./build.sh Codegen --configuration ${buildType}
'';
}
| 02:43:03 |
Whovian9369 | * Alternatively, is there a way to keep the in-progress /tmp/fetch-deps-... path even after a failed run? I feel like that would be good for testing.
The same process works fine on Net 8 but not Net 9, so I'm stumped from what I can tell.
{
preConfigure = ''
dotnet --version > DotnetCliVersion.txt
patchShebangs --build build.sh
for rid in $dotnetRuntimeIds; do dotnet restore --runtime "$rid" "build/_build.csproj"; done
./build.sh Codegen --configuration ${buildType}
'';
}
| 02:43:22 |
Corngood | In reply to @whovian9369:matrix.org
Alternatively, is there a way to keep the in-progress /tmp/fetch-deps-... path even after a failed run? I feel like that would be good for testing.
The same process works fine on Net 8 but not Net 9, so I'm stumped from what I can tell.
{
preConfigure = ''
dotnet --version > DotnetCliVersion.txt
patchShebangs --build build.sh
for rid in $dotnetRuntimeIds; do dotnet restore --runtime "$rid" "build/_build.csproj"; done
./build.sh Codegen --configuration ${buildType}
'';
}
There should be a fetch-drv attribute which you can run nix develop on to do the fetch in a dev shell. | 02:50:08 |
Whovian9369 | Oh thank you, I wasn't aware of that! | 02:59:19 |
Corngood | Not sure about the executable thing. It must be pulling that in from somewhere? | 03:03:23 |
Whovian9369 | If you mean the dynamic linked executable, then that's part of the Nuke build. | 03:07:21 |
Whovian9369 | I'm expecting the failing step to be the ./build.sh Codegen --configuration ${buildType} one from preConfigure that I posted above | 03:08:46 |
Whovian9369 | It works on Net 8 as expected, but trying the otherwise same derivation with the Net 9 SDK threw that error, for whatever reason. I'll probably have to try it with Net 8 (via fetch-drv?) and see how much the ldd output differs for these _build binaries.
Here's the Net 9 ldd on _build for example:
$ ldd _build
linux-vdso.so.1 (0x00007ffffa5f8000)
libpthread.so.0 => /nix/store/8p33is69mjdw3bi1wmi8v2zpsxir8nwd-glibc-2.40-66/lib/libpthread.so.0 (0x00007aef94629000)
libdl.so.2 => /nix/store/8p33is69mjdw3bi1wmi8v2zpsxir8nwd-glibc-2.40-66/lib/libdl.so.2 (0x00007aef94624000)
libstdc++.so.6 => not found
libm.so.6 => /nix/store/8p33is69mjdw3bi1wmi8v2zpsxir8nwd-glibc-2.40-66/lib/libm.so.6 (0x00007aef9453c000)
libgcc_s.so.1 => /nix/store/f6famnry0piih1d3hh6y73fxx05jxsa8-xgcc-14.3.0-libgcc/lib/libgcc_s.so.1 (0x00007aef9450e000)
libc.so.6 => /nix/store/8p33is69mjdw3bi1wmi8v2zpsxir8nwd-glibc-2.40-66/lib/libc.so.6 (0x00007aef94200000)
/lib64/ld-linux-x86-64.so.2 => /nix/store/8p33is69mjdw3bi1wmi8v2zpsxir8nwd-glibc-2.40-66/lib64/ld-linux-x86-64.so.2 (0x00007aef94645000)
Did MS or an Nixpkgs remove linking against libstdc between Net 8 and Net 9? That's the only issue that I can immediately find, for example...
| 03:11:48 |
Whovian9369 | * It works on Net 8 as expected, but trying the otherwise same derivation with the Net 9 SDK threw that error, for whatever reason. I'll probably have to try it with Net 8 (via fetch-drv?) and see how much the ldd output differs for these _build binaries.
Here's the Net 9 ldd on _build for example:
$ ldd _build
linux-vdso.so.1 (0x00007ffffa5f8000)
libpthread.so.0 => /nix/store/8p33is69mjdw3bi1wmi8v2zpsxir8nwd-glibc-2.40-66/lib/libpthread.so.0 (0x00007aef94629000)
libdl.so.2 => /nix/store/8p33is69mjdw3bi1wmi8v2zpsxir8nwd-glibc-2.40-66/lib/libdl.so.2 (0x00007aef94624000)
libstdc++.so.6 => not found
libm.so.6 => /nix/store/8p33is69mjdw3bi1wmi8v2zpsxir8nwd-glibc-2.40-66/lib/libm.so.6 (0x00007aef9453c000)
libgcc_s.so.1 => /nix/store/f6famnry0piih1d3hh6y73fxx05jxsa8-xgcc-14.3.0-libgcc/lib/libgcc_s.so.1 (0x00007aef9450e000)
libc.so.6 => /nix/store/8p33is69mjdw3bi1wmi8v2zpsxir8nwd-glibc-2.40-66/lib/libc.so.6 (0x00007aef94200000)
/lib64/ld-linux-x86-64.so.2 => /nix/store/8p33is69mjdw3bi1wmi8v2zpsxir8nwd-glibc-2.40-66/lib64/ld-linux-x86-64.so.2 (0x00007aef94645000)
Did MS or an Nixpkgs remove linking for libstdc++.so.6 between Net 8 and Net 9? That's the only issue that I can immediately find, for example...
| 03:13:57 |
Corngood | Nothing comes to mind. This executable must come from a nuget package? But how is it finding any of those libs? | 03:18:32 |
Whovian9369 |
This executable must come from a nuget package?
I believe the binary comes from for rid in $dotnetRuntimeIds; do dotnet restore --runtime "$rid" "build/_build.csproj"; done
But how is it finding any of those libs? Honestly, that's a good question that I don't have an answer to.
| 03:19:48 |
Whovian9369 | *
This executable must come from a nuget package?
I believe the binary comes from for rid in $dotnetRuntimeIds; do dotnet restore --runtime "$rid" "build/_build.csproj"; done
But how is it finding any of those libs?
Honestly, that's a good question that I don't have an answer to.
| 03:19:52 |
Corngood | If it went through patchelf I'd expect it to fail on missing dependencies. Maybe run `readelf -d` on it. | 03:21:32 |
Whovian9369 | $ readelf -d _build
Dynamic section at offset 0x106c8 contains 33 entries:
Tag Type Name/Value
0x000000000000001d (RUNPATH) Library runpath: [$ORIGIN/netcoredeps]
0x0000000000000001 (NEEDED) Shared library: [libpthread.so.0]
0x0000000000000001 (NEEDED) Shared library: [libdl.so.2]
0x0000000000000001 (NEEDED) Shared library: [libstdc++.so.6]
0x0000000000000001 (NEEDED) Shared library: [libm.so.6]
0x0000000000000001 (NEEDED) Shared library: [libgcc_s.so.1]
0x0000000000000001 (NEEDED) Shared library: [libc.so.6]
0x000000000000001e (FLAGS) BIND_NOW
0x000000006ffffffb (FLAGS_1) NOW PIE
0x0000000000000015 (DEBUG) 0x0
0x0000000000000007 (RELA) 0x1ba0
0x0000000000000008 (RELASZ) 744 (bytes)
0x0000000000000009 (RELAENT) 24 (bytes)
0x000000006ffffff9 (RELACOUNT) 20
0x0000000000000017 (JMPREL) 0x1e88
0x0000000000000002 (PLTRELSZ) 2448 (bytes)
0x0000000000000003 (PLTGOT) 0x12928
0x0000000000000014 (PLTREL) RELA
0x0000000000000006 (SYMTAB) 0x340
0x000000000000000b (SYMENT) 24 (bytes)
0x0000000000000005 (STRTAB) 0xfc4
0x000000000000000a (STRSZ) 3032 (bytes)
0x000000006ffffef5 (GNU_HASH) 0xfa8
0x0000000000000019 (INIT_ARRAY) 0x12630
0x000000000000001b (INIT_ARRAYSZ) 72 (bytes)
0x000000000000001a (FINI_ARRAY) 0x12628
0x000000000000001c (FINI_ARRAYSZ) 8 (bytes)
0x000000000000000c (INIT) 0x10f84
0x000000000000000d (FINI) 0x10fa0
0x000000006ffffff0 (VERSYM) 0xda8
0x000000006ffffffe (VERNEED) 0xe88
0x000000006fffffff (VERNEEDNUM) 5
0x0000000000000000 (NULL) 0x0
| 03:24:09 |
Corngood | Maybe you have LD_ vars or nix-ld?
In any case it's not patched, which is going to be a problem for building. | 03:28:57 |
Whovian9369 | (In case it matters, I used readelf from nixpkgs#llvmPackages_21.bintools) | 03:29:12 |
Corngood | Does build.sh pull in dependencies? Maybe it should be in preBuild. | 03:29:57 |
Corngood | Then it wouldn't get run in fetch, and maybe in a real build the executable would be patched | 03:30:22 |
Corngood | There are definitely projects using nuke in nixpkgs, and I don't remember them having this problem. | 03:31:35 |
Whovian9369 | I believe part of the issue there is that it's using Nuke for part of the build (for CodeGen) and Dotnet for the rest. | 03:32:44 |
Corngood | But why put it in preConfigure rather than preBuild. The restore stuff should stay in preConfigure | 03:35:09 |
Corngood | * But why put it in preConfigure rather than preBuild? The restore stuff should stay in preConfigure | 03:35:24 |
Whovian9369 | Can't say that I remember, but I'll see what I can mess with. Thanks! | 03:35:45 |
Whovian9369 | Here's what I'm trying out now:
preConfigure = ''
dotnet --version > DotnetCliVersion.txt
'';
preBuild = ''
patchShebangs --build build.sh
for rid in $dotnetRuntimeIds; do dotnet restore --runtime "$rid" "build/_build.csproj"; done
./build.sh Codegen --configuration ${buildType}
'';
| 03:36:50 |
Whovian9369 | That fixes it in fetch-deps at least, so let's see if a normal nix build succeeds here. | 03:37:15 |