From b61286004dd96d8774a5706ea25b72770ad43690 Mon Sep 17 00:00:00 2001 From: Alexandre Archambault Date: Tue, 28 Jun 2016 00:10:20 +0200 Subject: [PATCH] Add early internals document draft --- INTERNALS.md | 84 ++++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 84 insertions(+) create mode 100644 INTERNALS.md diff --git a/INTERNALS.md b/INTERNALS.md new file mode 100644 index 000000000..4cc41461c --- /dev/null +++ b/INTERNALS.md @@ -0,0 +1,84 @@ +# Internals / architecture + +## The model + +Mainly in [Definitions.scala](https://github.com/alexarchambault/coursier/blob/master/core/shared/src/main/scala/coursier/core/Definitions.scala), [Resolution.scala](https://github.com/alexarchambault/coursier/blob/master/core/shared/src/main/scala/coursier/core/Resolution.scala), and [ResolutionProcess.scala](https://github.com/alexarchambault/coursier/blob/master/core/shared/src/main/scala/coursier/core/ResolutionProcess.scala). + +### Module + +[definition](https://github.com/alexarchambault/coursier/blob/462c16d6db98d35e180a25b0f87aa47083ad98aa/core/shared/src/main/scala/coursier/core/Definitions.scala#L12-L16) + +Uniquely designates a... module. Typically, just an organisation (`org.scala-lang`) and a name (`scala-library`). At times, can also contains so called attributes: (unordered) key-value pairs. E.g. the SBT plugins usually have some, like `scalaVersion` with value `2.10` and `sbtVersion` with value `0.13`. + +Two modules having different organisation / name / attributes are simply considered different. + +During resolution, all dependencies with the same module have their versions reconciled if possible. If not possible, we have a version conflict (conflicting versions of a module are needed), which makes the resolution fail. + +### Dependency + +[definition](https://github.com/alexarchambault/coursier/blob/462c16d6db98d35e180a25b0f87aa47083ad98aa/core/shared/src/main/scala/coursier/core/Definitions.scala#L43-L54) + +A dependency towards a given module, with a given version (can be a version interval too, like `[2.2,2.3)` - TODO add support for `last.revision`). + +TODO Add a word about the various fields of `Dependency` + +### Project + +[definition](https://github.com/alexarchambault/coursier/blob/462c16d6db98d35e180a25b0f87aa47083ad98aa/core/shared/src/main/scala/coursier/core/Definitions.scala#L69-L96) + +Metadata about a given version of a module. + +Usually comes from a POM file (Maven) or an `ivy.xml` file (Ivy). In the former case, the infos in `Project` may originate from several POMs, if the main one has a parent POM, or some import dependencies. Raw `Project`s are obtained from the various individual POMs, and merged back into one (done during resolution by `Resolution` - TODO give more details). + +TODO Describe the various fields of `Project` + +### Resolution + +[definition](https://github.com/alexarchambault/coursier/blob/462c16d6db98d35e180a25b0f87aa47083ad98aa/core/shared/src/main/scala/coursier/core/Resolution.scala#L477-L487) + +State of the resolution. + +At any given point during resolution, we have: +- ... (TODO describe the various fields) + +Properties of Resolution: TODO describe the most important methods of `Resolution` + +### ResolutionProcess + +[definition](https://github.com/alexarchambault/coursier/blob/462c16d6db98d35e180a25b0f87aa47083ad98aa/core/shared/src/main/scala/coursier/core/ResolutionProcess.scala) + +Proceeds with the resolution per se, starting from an initial state (a `Resolution` instance), until the final one. + +Goes from one state to another by determining whether the current state: +- needs some extra metadata about some versions of some modules (some IO is needed, with a cache and/or repositories), or +- needs to take into account previously fed metadata (just calculations, no IO, giving the next state). + +So each step is either IO or calculations. Several IO steps can occur in a row, if some parent POMs or dependency imports are needed to get the full picture about a given module. A calculation step is either the last step, or followed by some IO. So the steps are like: IO, IO, IO, calculations, IO, IO, calculations, IO, calculations -> done. The last step is necessarily some calculations (IO can't end it - the newly fetched metadata of an IO step needs to be taken into account by a calculation step). + +TODO Describe ResolutionProcess a bit more... + +### Artifacts + +Once the resolution is done, we get the final `Resolution`. This `Resolution`can provide: +- final dependency list, +- artifact list, +- dependency graph, +- ... + +TODO Describe all of these a bit more + +## Cache / downloading + +### Fetching metadata + +What we need during resolution: fetching metadata, possibly several ones at once. + +TODO Describe the function to supply to `ResolutionProcess` to fetch metadata + +### Resolution result: artifact list + +TODO How we can fetch artifacts per se... + +### Cache structure on disk + +TODO More about the cache (structure on disk, error tracking, TTL tracking, locks)