[tracker/wip/carlosg/hotdoc: 8/47] docs: Port "performance" docs to markdown

[Carlos Garnacho <carlosg src gnome org>]

commit dd78ca9e52d316beb1679ba5fe33f7a8f23e3cc1
Author: Carlos Garnacho <carlosg gnome org>
Date:   Thu May 27 18:40:04 2021 +0200

    docs: Port "performance" docs to markdown

 docs/reference/libtracker-sparql/meson.build     |   1 +
 docs/reference/libtracker-sparql/performance.md  | 103 ++++++++++++++++
 docs/reference/libtracker-sparql/performance.xml | 142 -----------------------
 3 files changed, 104 insertions(+), 142 deletions(-)
---
diff --git a/docs/reference/libtracker-sparql/meson.build b/docs/reference/libtracker-sparql/meson.build
index 8e9f5b149..b528ac231 100644
--- a/docs/reference/libtracker-sparql/meson.build
+++ b/docs/reference/libtracker-sparql/meson.build
@@ -1,6 +1,7 @@
 content = [
   'overview.md',
   'limits.md',
+  'performance.md',
   'tutorial.md',
 ]
 
diff --git a/docs/reference/libtracker-sparql/performance.md b/docs/reference/libtracker-sparql/performance.md
new file mode 100644
index 000000000..7537657cb
--- /dev/null
+++ b/docs/reference/libtracker-sparql/performance.md
@@ -0,0 +1,103 @@
+Title: Performance dos and donts
+Slug: performance-dos-donts
+
+SPARQL is a very powerful query language. As it should be
+suspected, this means there are areas where performance is
+sacrificed for versatility.
+
+These are some tips to get the best of SPARQL as implemented
+by Tracker.
+
+## Avoid queries with unrestricted predicates
+
+Queries with unrestricted predicates are those like:
+
+```SPARQL
+SELECT ?p { <a> ?p 42 }
+```
+
+They involve lookups across all possible triples of
+an object, which roughly translates to a traversal
+through all tables and columns.
+
+The most pathological case is:
+
+```SPARQL
+SELECT ?s ?p ?o { ?s ?p ?o }
+```
+
+Which does retrieve every triple existing in the store.
+
+Queries with unrestricted predicates are most useful to
+introspect resources, or the triple store in its entirety.
+Production code should do this in rare occasions.
+
+## Avoid the negated property path
+
+The `!` negation operator in property paths negate the
+match. For example:
+
+```SPARQL
+SELECT ?s ?o { ?s !nie:url ?o }
+```
+
+This query looks for every other property that is not
+`nie:url`. The same reasoning than unrestricted predicates
+apply, since that specific query is equivalent to:
+
+```SPARQL
+SELECT ?s ?o {
+   ?s ?p ?o .
+   FILTER (?p != nie:url)
+}
+```
+
+## Specify graphs wherever possible
+
+Queries on the union graph, or with unrestricted graphs, for
+example:
+
+```SPARQL
+SELECT ?u { ?u a rdfs:Resource }
+SELECT ?g ?u { GRAPH ?g { ?u a rdfs:Resource }}
+```
+
+Will traverse across all graphs. Query complexity will increment
+linearly with the amount of graphs. Production code should rarely
+need to introspect graphs, and should strive to being aware of
+the graph(s) involved. The fastest case is accessing one graph.
+
+The graph(s) may be specified through
+`WITH / FROM / FROM NAMED / GRAPH` and other
+SPARQL syntax for graphs. For example:
+
+```SPARQL
+WITH <G> SELECT ?u { ?u a rdfs:Resource }
+WITH <G> SELECT ?g ?u { GRAPH ?g { ?u a rdfs:Resource }}
+```
+
+## Avoid substring matching
+
+Matching for regexp/glob/substrings defeats any index text fields
+could have. For example:
+
+```SPARQL
+SELECT ?u {
+  ?u nie:title ?title .
+  FILTER (CONTAINS (?title, "sideshow"))
+}
+```
+
+Will traverse all title strings looking for the substring. It is
+encouraged to use fulltext search for finding matches within strings
+where possible, for example:
+
+```SPARQL
+SELECT ?u { ?u fts:match "sideshow" }
+```
+
+## Use TrackerSparqlStatement
+
+Using [class@Tracker.SparqlStatement] allows to parse and compile
+a query once, and reuse it many times. Its usage
+is recommended wherever possible.