Tectonic is slower than xelatex #452
Description
This can be observed when generating a pdf for tests/xenia/paper.tex.
Using tectonic from master compiled with cargo install --path . and xelatex from Debian 9 in a i3-2330M 2.20GHz machine, I got the following results
> time tectonic -C -c minimal -r 1 paper.tex
real 0m1,642s
user 0m1,530s
sys 0m0,112s
> time xelatex -interaction=batchmode paper.tex
real 0m0,842s
user 0m0,796s
sys 0m0,120s
Running perf show that at least 25% of the time is expended by sha2 crate:
# Overhead Command Shared Object Symbol
# ........ ........ ........................ ...........................................................................................
#
23.78% tectonic tectonic [.] sha2::sha256_utils::compress256
10.66% tectonic tectonic [.] get_next
6.03% tectonic tectonic [.] load_fmt_file
3.61% tectonic tectonic [.] <std::io::buffered::BufReader<R> as std::io::Read>::read
3.55% tectonic tectonic [.] <tectonic::io::InputHandle as std::io::Read>::read
3.29% tectonic tectonic [.] sha2::sha256::Engine256::input
2.51% tectonic tectonic [.] macro_call
2.17% tectonic tectonic [.] length
1.76% tectonic [kernel.kallsyms] [k] clear_page_rep
1.71% tectonic tectonic [.] tectonic::io::InputHandle::getc
1.57% tectonic tectonic [.] flush_list
1.50% tectonic tectonic [.] get_uni_c
1.39% tectonic tectonic [.] tectonic::engines::input_getc
1.16% tectonic tectonic [.] id_lookup
1.16% tectonic tectonic [.] ht_lookup_table
Besides using hash to check if the engine needs to be rerun, what are the other uses? Is the hash of cached files checked? (I skimmed the code but was unable to answer that).