Add bindings to all jemalloc public functions

[SimonSapin]

That is, those documented in http://jemalloc.net/jemalloc.3.html

For example malloc and free can be useful when implementing callbacks for a C library that doesn’t provide a size when deallocating: https://www.freetype.org/freetype2/docs/reference/ft2-system_interface.html

[alexcrichton]

Thanks! The signature of sallocx may be off though?

[SimonSapin]

It’s quite possible since I translated from C declarations on the man page by hand, but this one seems to match?

    #[link_name = "_rjem_sallocx"]
    pub fn sallocx(ptr: *mut c_void, flags: c_int) -> size_t;

http://jemalloc.net/jemalloc.3.html

size_t sallocx( void *ptr,
                int flags);

[alexcrichton]

No idea myself, I'm just the messenger saying that CI is red

[SimonSapin]

This a warning-treated-as-error in C code that this PR didn’t modify. Maybe Travis updated their compiler version? Does this error happen on master if you hit the retry button?

[alexcrichton]

Mind looking into this? This is a test verifying the C API is correct. You added bindings for sallocx in this PR, and the C compiler is saying it's not correct.

[SimonSapin]

Oh sorry, I thought this C code was part of jemalloc, I didn’t realize it was generated from Rust declarations. Looking.

[SimonSapin]

Alright, pushed with a fix. The problem was that man page says void *ptr (which translates to *mut c_void) but the code says const void *ptr (which translates to *const c_void).

[alexcrichton]

I'm a tad bit wary about jemalloc::ffi, but perhaps instead of pub extern crate it could look like:

pub mod ffi {
    pub use jemalloc_sys::*;
}

?

[SimonSapin]

IIRC pub extern crate has been doing the same thing as that mod + pub use pattern since rust-lang/rust#31362.

[alexcrichton]

Ok... Can the version bumps be left out? I'm sort of wary with a release-per-PR, so maybe something like [patch] can be used to test these out and then they'll get published later?

[SimonSapin]

I’ve already made a Servo try build using these changes through [patch]. Everything is green other than a tidy failure in my Servo PR. In general we avoid landing usage of [patch] or [replace] in Servo master: this is effectively forking a library from upstream which is more of a last-resort option.

I can remove the version increment if you prefer. I’ve included it to signal that I’d appreciate a crates.io version including these changes to be published at your convenience. This doesn’t necessarily mean a separate version for every single PR, batching them is totally fine. A new version would enable this Servo PR to land as-is (with [patch] removed), and hopefully fix a segfault on Android. If you say that publishing needs to wait on something that’s likely to take time, I can look for alternatives.

[alexcrichton]

Nah it's fine to have, I just wanted to make sure that it was tested before release. So many PRs on other projects tend to have 3 successive releases in a row as they were never tested...

In any case, thanks again!