+gnu.org/glibc

[jhheider]

• debootstrap binutils/gcc
• build glibc
• use glibc without segfault

@mxcl // @jonchang I'd love some eyes on this. Currently, things using the new glibc like to segfault. I'm guessing I need to runtime.env.$some_env_vars to make it first on the list of libcs looked for.

[what-the-diff]

• binutils:
  • added 2.40 version, moved v2.39 to '2.39' (string)
  • removed bootstrap block for linux platform in build script; now depends on gcc/binutils package instead of installing via apt-get directly
  • changed test from objdump | grep $TEST_STRING to objdump | grep "$TEST_STRING" and also added a second line with the same command but using grep with the -s flag so that it will exit nonzero if no match is found

[jhheider]

I probably need to move the binutils/gcc stuff out to peel out the bootstraps.

[jhheider]

This passes testing, though I'm far from sure it's actually testing what I think it is. Next step is to make tea for linux selectively use it for tea-provided binaries iff the system's glibc is too old. 👀

[mxcl]

-nostdinc & -nostdlib

[jhheider]

that looks less promising. you sure we want both?

[mxcl]

it'll stop it using the system headers and libraries… which is surely what we want?

[jhheider]

Possibly -nodefaultlibs?

https://stackoverflow.com/questions/10119520/g-static-initialization-and-nostdlib

[mxcl]

Oh ok I see. gcc considers glibc’s headers and libraries the system libraries…?

The goal is for gcc to only use glibc and tea libs, nothing from /usr/lib or /lib or /include or /usr/include etc.

[jhheider]

which means, I think, providing the init loader from libgcc. we'll see.

[mxcl]

this pr has become epic in scope and effort

[felipecrs]

no wonder, it basically solves one of the world's biggest problems (i.e allows to run anything on alpine)

[jhheider]

Superseded by #12968.