Explicitly disable pie when pie is false

[yshui]

A lot of the platforms now has pie enabled by default, without any compiler flags. Maybe when the user explicitly set b_pie to be false, meson should explicitly disable pie with -no-pie.

[bwidawsk]

Is there any traction here?

Here is a test with pie: false

compile_commands.json: "command": "ccache cc -Ihello@exe -I. -I.. -fdiagnostics-color=always -pipe -D_FILE_OFFSET_BITS=64 -Wall -Winvalid-pch -g -MD -MQ 'hello@exe/main.c.o' -MF 'hello@exe/main.c.o.d' -o 'hello@exe/main.c.o' -c ../main.c",

And pie:true

"command": "ccache cc -Ihello@exe -I. -I.. -fdiagnostics-color=always -pipe -D_FILE_OFFSET_BITS=64 -Wall -Winvalid-pch -g -fPIE -MD -MQ 'hello@exe/main.c.o' -MF 'hello@exe/main.c.o.d' -o 'hello@exe/main.c.o' -c ../main.c",

The only way I can get a statically linked, PIE is with something like the following:

project('test', 'c')
hello = executable('hello',
  'main.c',
  c_args: [ '-ffreestanding', '-static'],
  link_args: ['-nostdlib', '-static', '-Wl,--entry=main'],
  pie: false
)

This leads me to the conclusion that pie as an argument is actually broken as the above is required to make pie: false work correctly, and with those link args and cflags, pie: true still generates a no-pic executable.

Tagging @marc-h38

[mstorsjo]

@dcbaker I see that you spent some effort on this early last year - any chance you might pick it up again? (I.e. +1, I also ran into this issue.)

[DidgeridooMH]

I also ran into this issue. What work is left to make this upstream?

[dcbaker]

I had patches that changed the behavior, but we wanted to make the behavior remain the same, ie, "nothing means add no flags and whatever the compiler defaults to" when you don't pass either true or false, and I never got around to fixing that. I've been in the midst of trying to make the interpreter argument handling more robust, and like several other changes here I've pushed this off until I can get that done, as it will make this work easier.

[dcbaker]

Also, we realized that pic has the same broken behavior, and needs to be fixed too.

[1ace]

I had patches that changed the behavior, but we wanted to make the behavior remain the same, [...], and I never got around to fixing that.

From what I can tell reading 9983262, it looks like you did do it 🙂

[marc-hb]

From closed #6415:

too many conflicts, too hard to rebase. I'll just rewrite from scratch

I haven't found any rewrite.

[dcbaker]

I have patches, but they’re blocked on #10714