Update HACKING and TODO
This commit is contained in:
parent
d475072b84
commit
9dfb78ea06
3 changed files with 80 additions and 36 deletions
79
HACKING.rst
79
HACKING.rst
|
@ -48,8 +48,8 @@ An artifact is an abstract representation of something that will produce output
|
|||
when it is written. It can be something like a file that will be copied
|
||||
verbatim, or a potential HTML page. Artifact can also be of higher order,
|
||||
meaning they wrap around other artifacts. The compound artifact is a simple
|
||||
case of just a plain container artifact. On the other hand, the blog artifact
|
||||
also has its own data such as the blog title.
|
||||
case of just a plain container artifact. On the other hand, the blog is a
|
||||
higher-order artifact that also has its own data such as the blog title.
|
||||
|
||||
Metadata
|
||||
========
|
||||
|
@ -87,6 +87,81 @@ Usually readers will be stored in a key-value variable where the key is the
|
|||
file format. Another function can then dispatch on the file format to the
|
||||
correct reader.
|
||||
|
||||
File systems and instructions
|
||||
=============================
|
||||
|
||||
Ultimately every artifact will be written to a file. Hard-coding this would
|
||||
require intermixing the concern of "represents some content" and "will produce
|
||||
a file". The lack of separation of concerns makes the artifact classes harder
|
||||
to reason about and harder to test. The act of producing an actual on-disk file
|
||||
needs to be separate from the individual artifacts. The usual solution would be
|
||||
to have some sort of "file system service" object which we can inject it as a
|
||||
dependency into a function, but this is not an elegant solution.
|
||||
|
||||
My solution are file systems and instructions. A file system is an object which
|
||||
abstracts away access to an actual file system. An instruction is an object
|
||||
which represents an action to perform, such as copying a file or writing a
|
||||
string to a file.
|
||||
|
||||
.. code:: lisp
|
||||
|
||||
(defclass base-file-system (file-system)
|
||||
((directory :initarg :directory :reader file-system-directory :type pathname
|
||||
:documentation "Actual directory within the file system."))
|
||||
(:documentation
|
||||
"A file system which accesses files of the host OS relative to a given base
|
||||
directory."))
|
||||
|
||||
(defclass write-string-contents (file-system-instruction)
|
||||
((path :initarg :path :reader instruction-path :type pathname
|
||||
:documentation "Path to the output file")
|
||||
(contents :initarg :contents :initform (make-string 0) :type string
|
||||
:documentation "The file content as a string"))
|
||||
(:documentation
|
||||
"Instruction which creates a file with the given content."))
|
||||
|
||||
We can apply an instruction to a file system, which carries out the action in
|
||||
the directory of the file system. We use the fact that CLOS supports multiple
|
||||
dispatch to dispatch on both the file system and the instruction.
|
||||
|
||||
.. code:: lisp
|
||||
|
||||
(defmethod write-to-filesystem ((instruction write-string-contents)
|
||||
(file-system base-file-system))
|
||||
"A primitive implementation producing one file for fixed contents and an
|
||||
absolute file system."
|
||||
(let ((path (fad:merge-pathnames-as-file
|
||||
(fad:pathname-as-directory (file-system-directory file-system))
|
||||
(instruction-path instruction))))
|
||||
(with-slots (contents) instruction
|
||||
(write-string-to-file contents path))))
|
||||
|
||||
But wait, if we have `n` file systems and `m` instructions, does that mean we
|
||||
need `n * m` implementations? No, most file systems and instructions are
|
||||
actually of a higher-order and reduce down to the most elemental ones. Consider
|
||||
the compound instruction, a container instruction which wraps other
|
||||
instructions:
|
||||
|
||||
.. code:: lisp
|
||||
|
||||
(defmethod write-to-filesystem ((instruction compound-instruction)
|
||||
file-system)
|
||||
(with-slots (instructions) instruction
|
||||
(dolist (instruction instructions)
|
||||
(write-to-filesystem instruction file-system))))
|
||||
|
||||
Instructions a produced by deriving an artifact; e.g. to derive an HTML
|
||||
artifact we generate the file contents as a string and produce a
|
||||
`WRITE-STRING-CONTENTS` instruction with the content and file name. We do not
|
||||
care where the file is written to, that part is the responsibility of the file
|
||||
system.
|
||||
|
||||
The most elemental file system simply references an on-disc directory. A
|
||||
higher-order file system is the overlay file system which adds a path on top of
|
||||
another file system. We are not bound by on-disc directories though: an FTP
|
||||
file system could abstract away access to an FTP server, a ZIP file system
|
||||
might abstract away access to a ZIP file.
|
||||
|
||||
|
||||
The blog plugin
|
||||
###############
|
||||
|
|
33
TODO.rst
33
TODO.rst
|
@ -8,36 +8,6 @@
|
|||
Core
|
||||
####
|
||||
|
||||
New feature: file systems
|
||||
=========================
|
||||
|
||||
Currently file output is strongly coupled to the file system of the OS. If we
|
||||
want to write an artifact, then writing the artifact is the responsibility of
|
||||
the artifact: it performs the low-level file system access, it generates the
|
||||
output text and it manages the file names, including the output directory path.
|
||||
|
||||
My proposal is to add a lever of indirection by separating concerns. There are
|
||||
three participants:
|
||||
|
||||
- File systems
|
||||
- Artifacts
|
||||
- Instructions
|
||||
|
||||
The artifact is an abstract representation of one or more future files. It is
|
||||
then *derived* to produce a low-level instruction on what action to actually
|
||||
perform to produce the file (relative file name, contents). The file system
|
||||
interprets the instruction by accessing the file systems and outputting the
|
||||
actual contents.
|
||||
|
||||
All this will be implemented using CLOS. A generic function dispatches on both
|
||||
instruction and file system. There will be core implementations for elemental
|
||||
instructions and file systems. Implementations for new classes will then be
|
||||
implemented on top of these primitive methods. For example, an implementation
|
||||
for an instruction which produces multiple files will created multiple
|
||||
lower-level instructions and call the generic function for each of these
|
||||
instructions and the original file system.
|
||||
|
||||
|
||||
New feature: sources
|
||||
====================
|
||||
|
||||
|
@ -71,9 +41,6 @@ Cleanup
|
|||
Testing
|
||||
=======
|
||||
|
||||
- Artifacts
|
||||
- Come up with a proper artifacts interface
|
||||
- Test the individual artifact implementations (function `WRITE-ARTIFACT`)
|
||||
- Update reader tests to public interface once it is done
|
||||
|
||||
|
||||
|
|
|
@ -61,7 +61,9 @@ directory."))
|
|||
;;; --- INSTRUCTION CLASSES ---------------------------------------------------
|
||||
(defclass write-string-contents (path-instruction)
|
||||
((contents :initarg :contents :initform (make-string 0) :type string
|
||||
:documentation "The file content as a string")))
|
||||
:documentation "The file content as a string"))
|
||||
(:documentation
|
||||
"Instruction which creates a file with the given content."))
|
||||
|
||||
(defclass copy-file (path-instruction)
|
||||
((base-path :initarg :base-path :type pathname
|
||||
|
|
Loading…
Add table
Reference in a new issue