Correct sysroot-prepend conditions for script inputs by parth-07 · Pull Request #809 · qualcomm/eld

parth-07 · 2026-02-11T14:18:08Z

This commit fixes the logic that determines when the script input path should be prepended with sysroot. The summary of the new behavior:

sysroot is only prepended to the script inputs when all the below constraints are true:
- the input is specified in INPUT/GROUP linker script command.
- The input path begins with '/'.
- The linker script that contains this INPUT/GROUP command is within the sysroot directory.
In all other cases, sysroot is not prepended to the script inputs.

Resolves #786
Resolves #808

quic-seaswara

Please update docs as well.

Please look into this change if verbose logging would be useful.

Map file diagnostics could be a follow up.

quic-seaswara · 2026-02-11T16:09:24Z

lib/Input/Input.cpp

+  Input *scriptInput = ParentScriptFile->getInput();
+
+  std::string scriptPath = scriptInput->getResolvedPath().getFullPath();
+  std::string sysrootPath = searchDirs.sysroot().getFullPath();


Will this work with reproduce option ?

Yes, I think so. We already had this code. Previously, we used to unconditionally prepend the sysroot for script inputs that began with /. Now, we only prepend sysroot for script inputs that begins with / if the parent script is within the sysroot directory.

Can you please check ?

I just checked. The linker crashes :'). However, this is not a regression with this patch. We crash without this patch as well.

Reproducer (nightly linker crashes with this):

#!/usr/bin/env bash mkdir -p A/lib64 mkdir -p A/script cat >1.c <<\EOF int foo() { return 1; } EOF cat >script.t <<\EOF GROUP(/lib64/lib1.so) EOF cat >main.c <<\EOF int main() { return 0; } EOF clang -o lib1.so -target x86_64-unknown-elf 1.c -shared -fPIC clang -o main.o -target x86_64-unknown-elf main.c -c mv lib1.so A/lib64/lib1.so mv script.t A/script.t ld.eld --sysroot=A -T A/script.t main.o --reproduce a.tar rm -rf unpacked && mkdir unpacked tar -xvf a.tar -C unpacked --strip-components=1 cd unpacked bash -x response.txt

I believe this should be treated as a separate issue.

I have created an issue to track this #818

lib/Input/Input.cpp

quic-seaswara · 2026-02-11T16:11:33Z

lib/Input/Input.cpp

+    return false;
+  if (Type != Input::Script)
+    return false;
+  if (FileName.empty() || FileName[0] != '/')


I am not sure how this will be effective on windows. Needs to be tested on windows too.

quic-seaswara · 2026-02-11T16:13:03Z

lib/Script/InputCmd.cpp

      break;
+    }
    case InputToken::NameSpec: {
      ThisBuilder.createDeferredInput(Token->name(), Input::Namespec);


Can we add the parent script file in all cases, and check if the command is INPUT or GROUP ?

Can you please explain this in more detail? Currently, we only add parent script for INPUT and GROUP inputs. If I understand correctly, then for the other cases, we do not have any associated parent script.

I see. Do we need to add this for InputToken::NameSpec too ?

No, we do not need this for NameSpec. But we do need to fix our behavior for NameSpec with sysroot. GNU LD finds the NameSpec library in all the standard paths inside sysroot. (Paths such as /usr/lib, /lib, /lib64, and so on).

Please open a follow up ticket on this, We do have this code in eld::ELDEmulateELF but that code is not getting triggered.

I have created an issue to track this: #819

parth-07 · 2026-02-11T17:32:46Z

Please update docs as well.

Is it okay if I push docs as follow-up? I am looking into a framework that will allow us to add extra information to https://qualcomm.github.io/eld/documentation/options/options.html. Currently, this page is automatically generated by parsing the tablegen option files.

Please look into this change if verbose logging would be useful.

Currently, we do have logs of the type Trying to open input file .... I think those logs are enough for this case as well. Please let me know your thoughts on this.

quic-seaswara · 2026-02-11T18:22:08Z

Please update docs as well.

Is it okay if I push docs as follow-up? I am looking into a framework that will allow us to add extra information to https://qualcomm.github.io/eld/documentation/options/options.html. Currently, this page is automatically generated by parsing the tablegen option files.

Please look into this change if verbose logging would be useful.

Currently, we do have logs of the type Trying to open input file .... I think those logs are enough for this case as well. Please let me know your thoughts on this.

Yes, this can be a follow up, but the document needs more than autogen treatment to explain the INPUT/GROUP commands appropriately.

quic-areg · 2026-02-12T00:08:53Z

lib/Input/Input.cpp

+  std::string scriptPath = scriptInput->getResolvedPath().getFullPath();
+  std::string sysrootPath = searchDirs.sysroot().getFullPath();
+
+  return scriptPath.size() >= sysrootPath.size() &&


what about a hypothetical /sdk/sysroot and a sibling path /sdk/sysroot_extra? we should probably use llvm::sys::fs::equivalent or similar instead of a raw prefix check

Here we just need to determine whether the script is within or outside the sysroot. If a script is contained within /sdk/sysroot_extra and the sysroot is /sdk/sysroot, then as per the linker the script is correctly outside the sysroot. The behavior seems correct to me. Can you please let me know if I have missed something?

This commit fixes the logic that determines when the script input path should be prepended with sysroot. The summary of the new behavior: - sysroot is only prepended to the script inputs when all the below constraints are true: - the input is specified in INPUT/GROUP linker script command. - The input path begins with '/'. - The linker script that contains this INPUT/GROUP command is within the sysroot directory. - In all other cases, sysroot is not prepended to the script inputs. Resolves qualcomm#786 Resolves qualcomm#808 Signed-off-by: Parth Arora <partaror@qti.qualcomm.com>

partaror requested review from quic-areg and quic-seaswara February 11, 2026 14:18

quic-seaswara requested changes Feb 11, 2026

View reviewed changes

parth-07 force-pushed the FixSysrootPrepend branch from d3d39f0 to 44c8f15 Compare February 11, 2026 17:35

parth-07 requested a review from quic-seaswara February 11, 2026 17:36

parth-07 force-pushed the FixSysrootPrepend branch from 44c8f15 to 9bbbb6f Compare February 11, 2026 17:46

parth-07 force-pushed the FixSysrootPrepend branch from 9bbbb6f to 711fb61 Compare February 11, 2026 18:26

quic-seaswara approved these changes Feb 11, 2026

View reviewed changes

jonathonpenix mentioned this pull request Feb 11, 2026

[cpullvm] Use eld throughout the Linux runtime build qualcomm/cpullvm-toolchain#134

Draft

quic-areg reviewed Feb 12, 2026

View reviewed changes

parth-07 force-pushed the FixSysrootPrepend branch from 711fb61 to ef8ca71 Compare February 12, 2026 11:41

Conversation

parth-07 commented Feb 11, 2026

Uh oh!

quic-seaswara left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

parth-07 Feb 12, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

parth-07 commented Feb 11, 2026

Uh oh!

quic-seaswara commented Feb 11, 2026

Uh oh!

quic-areg Feb 12, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

parth-07 Feb 12, 2026 •

edited

Loading

quic-areg Feb 12, 2026 •

edited

Loading