Optimise email CSS for speed with Premailer

Remove all descendant selectors from the push email styling, to
drastically reduce CPU time when inlining the CSS for syntax-highlighted
diffs.

Background:

Premailer is a Ruby gem that inlines CSS styles from an external
stylesheet before emails are sent, so that they are compatible with
Gmail. At a high level, it parses the CSS files it finds, and parses the
email body with Nokogiri. It then loops through the selectors in the
CSS, using Nokogiri to find matching elements, and adds inline
styles. (It does more than this, like merging styles applied to the same
element, but that's not relevant to this issue.)

Nokogiri converts CSS selectors to XPath first, like so:
    Nokogiri::CSS.xpath_for('foo bar')
    # => ["//foo//bar"]

On documents with high node counts (say, a syntax-highlighted copy of
jQuery), having both descendant selectors is very expensive. Both
`//foo/bar` and `//bar` will be much more efficient, although neither
are directly equivalent.

An example, on a document containing two syntax-highlighted copies of
jQuery:
    Benchmark.realtime { p doc.search('.o').count }
    # 9476
    # => 0.3462457580026239
    Benchmark.realtime { p doc.search('.code.white .o').count }
    # 9476
    # => 85.51952634402551

The performance is similar for selectors which _don't_ match any
elements, and as Premailer loops through all the available selectors, we
want to avoid all descendant selectors in push emails.

Because of the theming support in the web UI, all syntax highlighting
selectors are descendant selectors of classes like `.code.white` or
`.code.monokai`. There are over 60 CSS classes for syntax highlighting
styles alone, all of which are expressed in the inefficient form above.

In emails we always use the white theme, and were reusing the same CSS
file. But in emails, we don't need to descend from `.code.white` as that
will always be the theme, and we can also remove some other selectors
that are only applicable to the web UI. For the remaining descendant
selectors, we can convert them to child selectors, type selectors, or
class selectors as appropriate.

As in the example above, having no descendant selectors at all in the
push email CSS can provide a drastic (and surprising) performance
improvement.
This commit is contained in:
Sean McGivern 2016-06-06 15:14:36 +01:00
parent aa7cef9325
commit f4cedacc7b
2 changed files with 169 additions and 30 deletions

View file

@ -1,5 +1,15 @@
@import "framework/variables";
// This file is largely copied from `highlight/white.scss`, but modified to
// avoid all descendant selectors (`table td`). This is because the CSS inlining
// we use performs dramatically worse on descendant selectors than the
// alternatives.
// <https://gitlab.com/gitlab-org/gitlab-ee/issues/490#note_12283632>
//
// DO NOT ADD ANY DESCENDANT SELECTORS TO THIS FILE. Instead, use (in order of
// preference): plain class selectors, type (element name) selectors, or
// explicit child selectors.
table.code {
width: 100%;
font-family: monospace;
@ -11,33 +21,162 @@ table.code {
-premailer-cellspacing: 0;
-premailer-width: 100%;
td {
> tr > td {
line-height: $code_line_height;
font-family: monospace;
font-size: $code_font_size;
}
td.diff-line-num {
margin: 0;
padding: 0;
border: none;
background: $background-color;
color: rgba(0, 0, 0, 0.3);
padding: 0 5px;
border-right: 1px solid $border-color;
text-align: right;
min-width: 35px;
max-width: 50px;
width: 35px;
}
&.diff-line-num {
margin: 0;
padding: 0;
border: none;
padding: 0 5px;
border-right: 1px solid;
text-align: right;
min-width: 35px;
max-width: 50px;
width: 35px;
}
td.line_content {
display: block;
margin: 0;
padding: 0 0.5em;
border: none;
white-space: pre;
&.line_content {
display: block;
margin: 0;
padding: 0 0.5em;
border: none;
white-space: pre;
}
}
}
@import "highlight/white";
.line-numbers, .diff-line-num {
background-color: $background-color;
}
.diff-line-num, .diff-line-num a {
color: $black-transparent;
}
pre.code, .diff-line-num {
border-color: $table-border-gray;
}
.code.white, pre.code, .line_content {
background-color: #fff;
color: #333;
}
.diff-line-num {
&.old {
background-color: $line-number-old;
border-color: $line-removed-dark;
}
&.new {
background-color: $line-number-new;
border-color: $line-added-dark;
}
&.hll:not(.empty-cell) {
background-color: $line-number-select;
border-color: $line-select-yellow-dark;
}
}
.line_content {
&.old {
background-color: $line-removed;
> .line > span.idiff, > .line > span > span.idiff {
background-color: $line-removed-dark;
}
}
&.new {
background-color: $line-added;
> .line > span.idiff, > .line > span > span.idiff {
background-color: $line-added-dark;
}
}
&.match {
color: $black-transparent;
background-color: $match-line;
}
&.hll:not(.empty-cell) {
background-color: $line-select-yellow;
}
}
pre > .hll {
background-color: #f8eec7 !important;
}
span.highlight_word {
background-color: #fafe3d !important;
}
.hll { background-color: #f8f8f8 }
.c { color: #998; font-style: italic; }
.err { color: #a61717; background-color: #e3d2d2; }
.k { font-weight: bold; }
.o { font-weight: bold; }
.cm { color: #998; font-style: italic; }
.cp { color: #999; font-weight: bold; }
.c1 { color: #998; font-style: italic; }
.cs { color: #999; font-weight: bold; font-style: italic; }
.gd { color: #000; background-color: #fdd; }
.gd .x { color: #000; background-color: #faa; }
.ge { font-style: italic; }
.gr { color: #a00; }
.gh { color: #999; }
.gi { color: #000; background-color: #dfd; }
.gi .x { color: #000; background-color: #afa; }
.go { color: #888; }
.gp { color: #555; }
.gs { font-weight: bold; }
.gu { color: #800080; font-weight: bold; }
.gt { color: #a00; }
.kc { font-weight: bold; }
.kd { font-weight: bold; }
.kn { font-weight: bold; }
.kp { font-weight: bold; }
.kr { font-weight: bold; }
.kt { color: #458; font-weight: bold; }
.m { color: #099; }
.s { color: #d14; }
.n { color: #333; }
.na { color: teal; }
.nb { color: #0086b3; }
.nc { color: #458; font-weight: bold; }
.no { color: teal; }
.ni { color: purple; }
.ne { color: #900; font-weight: bold; }
.nf { color: #900; font-weight: bold; }
.nn { color: #555; }
.nt { color: navy; }
.nv { color: teal; }
.ow { font-weight: bold; }
.w { color: #bbb; }
.mf { color: #099; }
.mh { color: #099; }
.mi { color: #099; }
.mo { color: #099; }
.sb { color: #d14; }
.sc { color: #d14; }
.sd { color: #d14; }
.s2 { color: #d14; }
.se { color: #d14; }
.sh { color: #d14; }
.si { color: #d14; }
.sx { color: #d14; }
.sr { color: #009926; }
.s1 { color: #d14; }
.ss { color: #990073; }
.bp { color: #999; }
.vc { color: teal; }
.vg { color: teal; }
.vi { color: teal; }
.il { color: #099; }
.gc { color: #999; background-color: #eaf2f5; }

View file

@ -6,19 +6,19 @@ p.details {
font-style: italic;
color: #777
}
.footer p {
.footer > p {
font-size: small;
color: #777
}
pre.commit-message {
white-space: pre-wrap;
}
.file-stats a {
.file-stats > a {
text-decoration: none;
}
.file-stats .new-file {
color: #090;
}
.file-stats .deleted-file {
color: #b00;
> .new-file {
color: #090;
}
> .deleted-file {
color: #b00;
}
}