rustpkg may be ignoring -O flag

[catamorphism]

@chrismorgan reports that rustpkg install -O http (after git cloneing http://github.com/chris-morgan/rust-http ) seems to silently ignore the -O flag. (I'm not yet sure whether this is because it should be rustpkg -O install http or not, as I haven't reproduced it yet. Of course, both ways should work.)

[chris-morgan]

I'm getting the same results (viz. no optimisations) for -O before or after the subcommand and for both the install and build subcommands.

Here are rust-http sizes, for example:

• rustc -O: 1244971
• rustpkg (-O or not): 1710369 or 1710377, as measured (don't ask me, I'm just the messenger who ran rustpkg build http multiple times after cleaning and got different results)
• rustc: 1710145 (why is this not bitwise identical to the rustpkg build?)

[catamorphism]

On IRC, @huonw noted that even rustc itself doesn't always generate an identical binary for identical input. (Though that's tangential to the issue with the -O flag.)

[huonw]

(It just appears that the binary from crates without a #[link(name)] depends on the value of -o on Linux. That is, compiling multiple times with the same -o or adding the name attribute gives the same result each time.)

[itdaniher]

In the meantime, you might try seeing if rustpkg --opt-level=3 build package yields different results, as a workaround.

No dice.

Anyone have other ideas for a workaround?

I've been continuing to poke around the rustpkg source, but I can't seem to build it without rebuilding the whole tree - rustpkg build librustpkg is a nogo.

[emberian]

#10526 supposedly fixed, just needs a test.

[darnuria]

Rustpkg was dropped... Transfers the bug to the actual repository of rustpkg and close it?