1 files changed, 100 insertions, 0 deletions
diff --git a/content/blog/hugo/search.md b/content/blog/hugo/search.md
new file mode 100644
index 0000000..d58f51c
--- /dev/null
+++ b/content/blog/hugo/search.md
@@ -0,0 +1,100 @@
+---
+title: Implementing a search feature for my hugo static website
+description: A golang webservice to complement hugo
+date: 2021-09-19
+tags:
+- golang
+- hugo
+---
+
+## Introduction
+
+When I migrated my personal website from dokuwiki to hugo two years ago, the main feature I lost was a way to search the website content. I used [tags](/tags) to mimic a fraction of this power, but with more and more articles over the years I felt the pressure to finally do something.
+
+Hugo can easily generate a json index of the website, and according to my google-fu hugo users use javascript solutions to implement search on top of this. I was not satisfied by the idea of having javascript download the whole website index and running searches locally, but I found no alternative. Since I love having a javascript free website I wanted to keep it that way if possible, so I designed an alternative.
+
+This blog post retraces my steps in programming a custom search feature in golang and hosting it alongside my blog on my personal infrastructure.
+
+## The design
+
+Up until now the website was served from a simple nginx configuration. I plan to write a golang webservice, have nginx redirect the `/search` path to it and keep serving all the other paths statically. Having some experience from my work on my [trains webapp](https://git.adyxax.org/adyxax/trains) I had a pretty good idea how to write the http handler, I just needed to figure out three things :
+- How to generate the JSON index with the information I need.
+- How to make the search page integrate perfectly with the website.
+- How to leverage the JSON index and which search algorithm to implement.
+
+### The JSON index
+
+There are two things to configure in order to get a JSON index from hugo. The first one is to activate JSON as an output format in `config.toml` :
+```toml
+[outputs]
+home = ["HTML", "RSS", "JSON"]
+```
+
+The second is to create a `layouts/index.json` template of what the index will contain. Here is mine, if you decide to tweak it you will need to rework the search algorithm accordingly :
+```html
+{{- $.Scratch.Add "index" slice -}}
+{{- range .Site.RegularPages -}}
+    {{- $.Scratch.Add "index" (dict "title" .Title "description" .Params.description "tags"
+    .Params.tags "content" .Plain "permalink" .Permalink) -}}
+{{- end -}}
+{{- $.Scratch.Get "index" | jsonify -}}
+```
+
+### How to maintain a coherent look
+
+Since I am writing a golang webservice I can use its HTML templating library. It is handy because hugo uses the same library, but I cannot just reuse my hugo templates since they use hugo functions and api which would be a headache to mimic. One of the main pain points is to be able to reuse the css hugo generates, fingerprint and all.
+
+In order to solve this I thought about my [custom hugo shortcodes]({{< ref "adding-custom-shortcode-age" >}}) and figured a way to use it to my advantage. I added a content section with a single page in `content/search/_index.md` :
+```markdown
+---
+title: "Search"
+menu:
+  main:
+    weight: 1
+layout: single
+---
+
+What are you looking for ?
+```
+
+To render this page I wrote the corresponding layout in `layouts/search/single.html` :
+```html
+{{ define "main" }}
+
+{{ .Content }}
+
+{{ print "{{ template \"search\" . }}" }}
+
+{{ end }}
+```
+
+And VOILÀ! When building the hugo website, it generates a page that contains a simple and valid html template that I can render :
+```html
+<!-- headers with css and everything exactly as on every page of the website,
+     as well as the menu and its section highlighting) -->
+<p>What are you looking for ?</p>
+{{ template "search" . }}
+<!-- the footer -->
+```
+
+The `search` template just need to be written accordingly, and the http templating library will do the rest.
+
+### The golang webservice
+
+The webservice lives in a folder of my hugo repository and can be found [here](https://git.adyxax.org/adyxax/www/src/branch/master/search). The website's makefile first builds the hugo website, then copies the HTML template and the json index in the search folder. It then builds the golang binary and embeds these.
+
+When the webservice starts, it parses the JSON index and generates separate lists of unique words found in titles, descriptions, tags and page content. These lists each have a weight that factors in the results when the searched words are found in the list via a simple `string.Contains` match.
+
+It is not very clever but because my website is small it produces pertinent results with this very simple algorithm. A friend pointed to me that Radix Trees are the efficient way to search if a word is part of a text and I plan to look into implementing that one day.
+
+## Conclusion
+
+I find that this search feature works well and looks even better! Being light and fast it fits perfectly with my philosophy for this website, and I find that managing a server side service is preferable to pushing the burden to the visitor's browser.
+
+Future improvements I am considering (beside radix trees) are :
+- allow to search for a phrase, right now words are split and handled similarly as if a OR was used.
+- allow to request a word be present in the results (right now it is a OR).
+- allow to match negatively against a word or pattern.
+- permit a verbatim search to prevent substring matches.
+- optionally select tags from a list, either positively or negatively.
+- maybe factor pages' publishing date as a results sorting option.