I did a few experiments regarding the binary size of a very small program. Initially I was worried how much space the symbols of libunwind occupied when analyzed with bloaty. But the results are interesting in other ways and I don’t understand them entierly.
The Program:
fn main() {
println!("Hello World!");
panic!();
}
And the profile in Cargo.toml
[profile.release]
panic = "abort"
lto = true
codegen-units = 1
incremental = false
[profile.dev]
panic = "abort"
I compiled the program in four configurations: stable and nightly installed using rustup. Then I compiled rust myself with the default config.toml -> backtrace. After that I set backtrace=false in config.toml -> nobacktrace
| debug | debug & stripped | release | release & stripped | |
|---|---|---|---|---|
| stable | 3,9M | 419K | 2,0M | 387K |
| nightly | 2,4M | 195K | 845K | 167K |
| backtrace | 308K | 195K | 194K | 159K |
| nobacktrace | 209K | 111K | 105K | 79K |
If you want to have a look yourself, here are the binaries: https://www.file-upload.net/download-13447445/binaries.tar.gz.html.
- In stable I think a substantial part of the extra size results from jemalloc.
- nightly and backtrace should have a similar size because they have similar source versions. For the stripped binaries this is the case, but there seems to be a lot of extra debug information in the program compiled with the nightly version. Is this intentional and what is it good for?
I really like the size reduction when compiled with backtrace=false. Maybe it makes sense to give this option without recompilation.
What do you think?