Performance when compiled under debug about handlebars-rust

[sunng87]

Could you please show me the template so I can reproduce this issue?

Does it have complex nested handlebars structure? Handlebars doesn't try to understand xml tags and it treats all these xml as plain text.

[mikedilger]

This was the file (when unzipped). It actually has no handlebars substitutions/commands in it:
pi.zip

This is quite low priority, just a curiosity. I've changed my code to (1) substitute the SVG via handlebars, rather than have it in the template, and (2) re-register templates only if their mtime has changed. So I'm not affected by this any more.

[sunng87]

Thanks for reporting this issue. I will look into this later, maybe in the weekend.

Perhaps I will need to find some tool to profile rust code. That will be pretty interesting.

[TimNN]

I did a bit of profiling and found the culprit:

Template::compile is at least O(n^2) (n is the input length) with a large constant factor: It repeatedly indexes the template string by character which is an O(n) operation.

Edit: And while doing so it decodes every character from UTF-8 to char.

[sunng87]

I can see while I have a counter scanning over the template string, I use several slice_char_alt which will scan the string from the start. That will lead to O(n^2).

Perhaps I can use some vector API over the chars array.

[TimNN]

It may be possible to do most of the parsing based on the bytes in the string, using b"byte literals" and counting bytes, because indexing a string by byte is O(1).

[sunng87]

According to my profiling result:

opannotate (warning): unable to open for reading: /home/nsun/projects/handlebars/../src/libcore/str/mod.rs
/* 
 * Total samples for file : "/home/nsun/projects/handlebars/../src/libcore/str/mod.rs"
 * 
 * 872171 41.2792
 */


 /* str::next_code_point::ha0fdfe659d904be7O1R total: 447023 21.1573 */
 /* str::Chars$LT$$u27$a$GT$.Iterator::next::h5ef996abf4a9cfe7D5R total: 120144  5.6863 */
 /* core::str::Chars$LT$$u27$a$GT$.Iterator::next::_$LT$closure$GT$::closure.13081 total: 147425  6.9775 */
 /* str::CharIndices$LT$$u27$a$GT$.Iterator::next::h9996f6bfe21bd4d5c8R total: 157076  7.4343 */
 /* str::eq_slice::h41e3b146382d4d2frFS total:    162  0.0077 */
 /* str::cmp_slice::hf5fdb32ea40fa722TFS total:     63  0.0030 */
 /* str::traits::str.PartialEq::eq::hd016519747f5b7cfsTS total:      2 9.5e-05 */
 /* str::traits::str.ops..Index$LT$ops..RangeFull$GT$::index::ha0436e16eb25ba6aTYS total:     17 8.0e-04 */
 /* str::str.StrExt::chars::hd207ee132a42af89v9S total:     18 8.5e-04 */
 /* str::str.StrExt::char_indices::h5b3c35a89f19d245T9S total:     12 5.7e-04 */
 /* str::str.StrExt::slice_unchecked::h755d1aaf0a77fde8IeT total:     59  0.0028 */
 /* str::str.StrExt::as_bytes::hc7a65f1a85e265b5gnT total:     82  0.0039 */
 /* str::str.StrExt::as_ptr::hc35e1d64caf0535fWqT total:     40  0.0019 */
 /* str::str.StrExt::len::h2b44ab638608b9be7qT total:     48  0.0023 */
opannotate (warning): unable to open for reading: /home/nsun/projects/handlebars/../src/libcore/slice.rs
/* 
 * Total samples for file : "/home/nsun/projects/handlebars/../src/libcore/slice.rs"
 * 
 * 523502 24.7770
 */


 /* slice::_$u5b$T$u5d$.SliceExt::iter::h13192447119355866019 total:     36  0.0017 */
 /* slice::_$u5b$T$u5d$.SliceExt::get_unchecked::h4309850808731124342 total:      1 4.7e-05 */
 /* slice::_$u5b$T$u5d$.SliceExt::get_unchecked::h6289303913257139361 total:     20 9.5e-04 */
 /* slice::_$u5b$T$u5d$.SliceExt::as_ptr::h12716624842058259554 total:     15 7.1e-04 */
 /* slice::_$u5b$T$u5d$.SliceExt::len::h15903203773402614389 total:     55  0.0026 */
 /* slice::_$u5b$T$u5d$.SliceExt::get_unchecked_mut::h10282977931327093713 total:     14 6.6e-04 */
 /* slice::_$u5b$T$u5d$.SliceExt::as_mut_ptr::h15680318839540136072 total:      5 2.4e-04 */
 /* slice::Iter$LT$$u27$a$C$$u20$T$GT$.Iterator::next::h5032969495063014924 total: 182078  8.6176 */
 /* slice::Iter$LT$$u27$a$C$$u20$T$GT$.Iterator::size_hint::h18247376353165846459 total: 341261 16.1516 */
 /* slice::from_raw_parts::h13925463592529871622 total:      7 3.3e-04 */
 /* slice::from_raw_parts_mut::h11037529489009763374 total:     10 4.7e-04 */

Most time are spent on string operations. Let me look into this issue later.

[sunng87]

@mikedilger the performance issue should have been fixed in 0.13.2