There are some breaking changes in this commit, see #11455.
Closes#11455Closes#11549
This fixes a set of bugs (see issue list) and it is also paying some technical debt accumulated over the years. We now build with Staticcheck enabled in the CI build.
The performance should be about the same as before for regular sized Hugo sites, but it should perform and scale much better to larger data sets, as objects that uses lots of memory (e.g. rendered Markdown, big JSON files read into maps with transform.Unmarshal etc.) will now get automatically garbage collected if needed. Performance on partial rebuilds when running the server in fast render mode should be the same, but the change detection should be much more accurate.
A list of the notable new features:
* A new dependency tracker that covers (almost) all of Hugo's API and is used to do fine grained partial rebuilds when running the server.
* A new and simpler tree document store which allows fast lookups and prefix-walking in all dimensions (e.g. language) concurrently.
* You can now configure an upper memory limit allowing for much larger data sets and/or running on lower specced PCs.
We have lifted the "no resources in sub folders" restriction for branch bundles (e.g. sections).
Memory Limit
* Hugos will, by default, set aside a quarter of the total system memory, but you can set this via the OS environment variable HUGO_MEMORYLIMIT (in gigabytes). This is backed by a partitioned LRU cache used throughout Hugo. A cache that gets dynamically resized in low memory situations, allowing Go's Garbage Collector to free the memory.
New Dependency Tracker: Hugo has had a rule based coarse grained approach to server rebuilds that has worked mostly pretty well, but there have been some surprises (e.g. stale content). This is now revamped with a new dependency tracker that can quickly calculate the delta given a changed resource (e.g. a content file, template, JS file etc.). This handles transitive relations, e.g. $page -> js.Build -> JS import, or $page1.Content -> render hook -> site.GetPage -> $page2.Title, or $page1.Content -> shortcode -> partial -> site.RegularPages -> $page2.Content -> shortcode ..., and should also handle changes to aggregated values (e.g. site.Lastmod) effectively.
This covers all of Hugo's API with 2 known exceptions (a list that may not be fully exhaustive):
Changes to files loaded with template func os.ReadFile may not be handled correctly. We recommend loading resources with resources.Get
Changes to Hugo objects (e.g. Page) passed in the template context to lang.Translate may not be detected correctly. We recommend having simple i18n templates without too much data context passed in other than simple types such as strings and numbers.
Note that the cachebuster configuration (when A changes then rebuild B) works well with the above, but we recommend that you revise that configuration, as it in most situations should not be needed. One example where it is still needed is with TailwindCSS and using changes to hugo_stats.json to trigger new CSS rebuilds.
Document Store: Previously, a little simplified, we split the document store (where we store pages and resources) in a tree per language. This worked pretty well, but the structure made some operations harder than they needed to be. We have now restructured it into one Radix tree for all languages. Internally the language is considered to be a dimension of that tree, and the tree can be viewed in all dimensions concurrently. This makes some operations re. language simpler (e.g. finding translations is just a slice range), but the idea is that it should also be relatively inexpensive to add more dimensions if needed (e.g. role).
Fixes#10169Fixes#10364Fixes#10482Fixes#10630Fixes#10656Fixes#10694Fixes#10918Fixes#11262Fixes#11439Fixes#11453Fixes#11457Fixes#11466Fixes#11540Fixes#11551Fixes#11556Fixes#11654Fixes#11661Fixes#11663Fixes#11664Fixes#11669Fixes#11671Fixes#11807Fixes#11808Fixes#11809Fixes#11815Fixes#11840Fixes#11853Fixes#11860Fixes#11883Fixes#11904Fixes#7388Fixes#7425Fixes#7436Fixes#7544Fixes#7882Fixes#7960Fixes#8255Fixes#8307Fixes#8863Fixes#8927Fixes#9192Fixes#9324
The change in lock logic for `partialCached` in 0927cf739f was naive as it didn't consider cached partials calling other cached partials.
This changeset may look on the large side for this particular issue, but it pulls in part of a working branch, introducing `context.Context` in the template execution.
Note that the context is only partially implemented in this PR, but the upcoming use cases will, as one example, include having access to the top "dot" (e.g. `Page`) all the way down into partials and shortcodes etc.
The earlier benchmarks rerun against master:
```bash
name old time/op new time/op delta
IncludeCached-10 13.6ms ± 2% 13.8ms ± 1% ~ (p=0.343 n=4+4)
name old alloc/op new alloc/op delta
IncludeCached-10 5.30MB ± 0% 5.35MB ± 0% +0.96% (p=0.029 n=4+4)
name old allocs/op new allocs/op delta
IncludeCached-10 74.7k ± 0% 75.3k ± 0% +0.77% (p=0.029 n=4+4)
```
Fixes#9519
This commit also
* revises the change detection for templates used by content files in server mode.
* Adds a Page.RenderString method
Fixes#6545Fixes#4663Closes#6043
This is a big commit, but it deletes lots of code and simplifies a lot.
* Resolving the template funcs at execution time means we don't have to create template clones per site
* Having a custom map resolver means that we can remove the AST lower case transformation for the special lower case Params map
Not only is the above easier to reason about, it's also faster, especially if you have more than one language, as in the benchmark below:
```
name old time/op new time/op delta
SiteNew/Deep_content_tree-16 53.7ms ± 0% 48.1ms ± 2% -10.38% (p=0.029 n=4+4)
name old alloc/op new alloc/op delta
SiteNew/Deep_content_tree-16 41.0MB ± 0% 36.8MB ± 0% -10.26% (p=0.029 n=4+4)
name old allocs/op new allocs/op delta
SiteNew/Deep_content_tree-16 481k ± 0% 410k ± 0% -14.66% (p=0.029 n=4+4)
```
This should be even better if you also have lots of templates.
Closes#6594
