Question about the build / package of libpdfium.so for new arch [ppc64le]

[mgiessing]

I've successfully built libpdfium.so for ppc64le architecture, but when I tried to package it into a pypdfium2 manylinux wheel (2014/glibc2.17+ & 2_28/glibc2.28+) I figured out there must be an issue somewhere as I got:

[...]
OSError: /lib64/libm.so.6: version `GLIBC_2.27' not found (required by /pypdfium2/data/linux_ppc64le/libpdfium.so)
[...]

I then inspected the libpdfium for aarch64, x86_64 & ppc64le and saw that for the first two the highest glibc was indeed 2.17, but for ppc64le I had up to 2.29:

aarch64:

$ objdump -x lib/libpdfium.so | grep GLIBC
    0x06969197 0x00 05 GLIBC_2.17
    0x06969197 0x00 04 GLIBC_2.17
    0x06969197 0x00 02 GLIBC_2.17
    0x06969197 0x00 03 GLIBC_2.17

x86_64:

$ objdump -x lib/libpdfium.so | grep GLIBC
    0x09691a75 0x00 08 GLIBC_2.2.5
    0x09691972 0x00 11 GLIBC_2.3.2
    0x09691a75 0x00 06 GLIBC_2.2.5
    0x09691a75 0x00 02 GLIBC_2.2.5
    0x0d696913 0x00 03 GLIBC_2.3
    0x09691973 0x00 14 GLIBC_2.3.3
    0x09691974 0x00 05 GLIBC_2.3.4
    0x0d696914 0x00 04 GLIBC_2.4
    0x0d696917 0x00 12 GLIBC_2.7
    0x06969191 0x00 09 GLIBC_2.11
    0x06969194 0x00 07 GLIBC_2.14
    0x06969196 0x00 15 GLIBC_2.16
    0x0d696913 0x00 10 GLIBC_2.3

ppc64le:

$ objdump -x lib/libpdfium.so | grep GLIBC
    0x06969197 0x00 04 GLIBC_2.17
    0x06969197 0x00 03 GLIBC_2.17
    0x06969187 0x00 05 GLIBC_2.27
    0x06969189 0x00 06 GLIBC_2.29
    0x06969197 0x00 02 GLIBC_2.17
    0x06969197 0x00 07 GLIBC_2.17

I did 3 main changes to make it work:

• Built debian_bullseye_linux_ppc64el with the sysroot_creator.py script
• Built libclang_rt.builtins.a for powerpc64le using the clang build.py script
• Added ppc64le to various places in the build toolchain

I believe the first two should be fine as I did it in the same git checkout but I'm wondering if I have somewhere forget a spot in the build toolchain to restrict the lib to e.g. 2.17?

Btw. I tried to keep it as simple as possible, here is the patch I apply after checking out:
https://github.com/mgiessing/pdfium-binaries/blob/ppc64le/patches/build_config_ppc64le.patch

Any help is appreciated :)

[mgiessing]

Based on this issue #82 I saw in my sysroot.../libc6/DEBIAN/symbols I also got too recent symbols for 2.27 & 2.29 - I just don't know how to proper process of creating this looks like.
When I build a new arm64 sysroot using the sysroot_creator.py script it also has newer symbols (2.27, 2.29...) in it - so I guess there is a different way of creating that image...

I was more or less following the chromium notes here

Again, any help is appreciated :)

[bblanchon]

Hi @mgiessing,

Congratulations on building PDFium no ppc64le 🎉

Does it need the host to be ppc64le, or can it be cross-built?
If we want to include this in the weekly build, it must build on an x64 host.

I wish I could help you, but as you saw in #82, I'm clueless about the GLIB version issues.
Your best chance is to ask on the PDFium group.

Best regards,
Benoit

[mgiessing]

Thank you! It can be cross-compiled on an x86_64 system - I have a AlmaLinux8.9 and tried it manually using:

dnf install -y git podman
mkdir -p $HOME/git && cd $HOME/git

git clone -b ppc64le https://github.com/mgiessing/pdfium-binaries 

# pdfium-binaries use ubuntu:24.04 in their GH Actions
podman run -ti -v $HOME/git:/root/git --cap-add=AUDIT_WRITE ubuntu:24.04
apt update && apt install sudo lsb-release file git python3 python3-pip wget curl -y
cd /root/git/pdfium-binaries/
PDFium_BRANCH="chromium/6954" bash build.sh linux ppc64le

However there are two "dirty" things that should probably be changed to the upstream pdfium/build/third_party project:

• They don't build debian_bullseye_ppc64el_sysroot, that's why I built it manually, uploaded it and patched into the sysroots.json:

+    },
+    "bullseye_ppc64el": {
+        "Sha256Sum": "d2e359f5460d2a70fb938c91139b8022fe4d85475567baaa7864eb39073f4c5a",
+        "SysrootDir": "debian_bullseye_ppc64el-sysroot",
+        "Tarball": "debian_bullseye_ppc64el_sysroot.tar.xz",
+        "URL": "https://ibm.box.com/shared/static/n9n2zcuybktrgbtph6q6bsif02pkpfak.xz"

• Same for libclang - I built & uploaded the libclang_rt.builtins and get them after checkout:

#Get libclang-rt for powerpc64le
pushd "${SOURCE}/third_party/llvm-build/Release+Asserts/lib/clang/20/lib/"
wget https://ibm.box.com/shared/static/7ez5z7vu75c0u2dno5t3anszwd2wrgdv.gz -O clang-rt-20-ppc.tar.gz
mkdir -p powerpc64le-unknown-linux-gnu
tar -xf clang-rt-20-ppc.tar.gz
rm -rf clang-rt-20-ppc.tar.gz
popd

Once all of this matured better I'm happy to provide a clean PR :)