.. rst3: filename: html/appendices.html ########## Appendices ########## .. |br| raw:: html
.. contents:: Contents :depth: 2 :local: Format of .leo files ++++++++++++++++++++ .. _`Writing plugins`: writingPlugins.html Here are the XML elements that may appear in Leo files: ```` Leo files start with the following line:: ```` This element is optional. For example:: ```` This element opens an element that contains the entire file. ```` ends the file. ```` This element specifies version information and other information that affects how Leo parses the file. For example:: The ``file_format`` attribute gives the 'major' format number. It is ``"2"`` for all newer versions of Leo. ````, ````, ```` and ```` These elements are vestigial. |br| Modern versions of Leo ignore them when reading outlines and write empty elements when writing outlines. ```` A single ```` element contains nested ```` elements. ```` elements correspond to vnodes. The nesting of ```` elements indicates outline structure in the obvious way. ```` This element represents a single vnode and has the following form:: sss (zero or more nested v elements) The ```` element specifies the headline text. sss is the headline text encoded with the usual XML escapes. As shown above, a ```` element may contain nested ```` elements. Zero or more of the following attributes may appear in elements:: t=name.timestamp.n a="xxx" The ``t="Tnnn"`` attribute specifies the element associated with a element. The ``a="xxx"`` attribute specifies vnode attributes. ``xxx`` denotes one or more upper-case letters whose meanings are as follows:: C The vnode is a clone. (Not used in 4.x) E The vnode is expanded so its children are visible. M The vnode is marked. T The vnode is the top visible node. V The vnode is the current vnode. For example, ``a="EM"`` specifies that the vnode is expanded and is marked. .. The tnode list existed from Leo 4.0 until Leo 4.7. Leo 4.7 removed the tnode list. - Plugins and scripts may add attributes to and elements. See `Writing plugins`_ for details. ```` A single ```` element contains a non-nested list of ```` elements. ```` This element represents the body text of the corresponding element. It has this form:: sss The ``tx`` attribute is required. The ``t`` attribute of ```` elements refer to this ``tx`` attribute. ``sss`` is the body text encoded with the usual XML escapes. **New in 4.0**: Plugins and scripts may add attributes to ```` and ```` elements. See `Writing plugins`_ for details. Format of external files ++++++++++++++++++++++++ .. _`sentinel lines`: glossary.html#sentinel-lines This section describes Leo's `sentinel lines`_, comment used in external files. .. index:: gnx External files created by ``@file`` use gnx's in ``@+node`` sentinels. Such gnx's permanently and uniquely identify nodes. Gnx's have the form:: id.yyyymmddhhmmss id.yyyymmddhhmmss.n The second form is used if two gnx's would otherwise be identical. - ``id`` is a string unique to a developer, e.g., a git id. - ``yyyymmddhhmmss`` is the node's creation date. - ``n`` is an integer. Closing sentinels are required for section references and the ``@all`` and ``@others`` directives, collectively known as **embedding constructs.** Proof: These constructs do not terminate the node in which they appear. Without a closing sentinel there would be no way to know where the construct ended and the following lines of the enclosing node began. Here are the sentinels used by modern versions of Leo, in alphabetical order. ``@<<`` Sentinels of the form ``@<>`` represent section references. If the reference does not end the line, the sentinel line ending the expansion is followed by the remainder of the reference line. This allows the Read code to recreate the reference line exactly. ``@@`` This sentinel represents any line starting with ``@`` in body text except ``@``, ``@doc`` and ``@others``. Examples:: @@nocolor @@pagewidth 80 @@tabwidth -4 @@c .. ``@afterref`` .. Marks non-whitespace text appearing after a section references. .. As of Leo 6.6, Leo no longer *writes* afterref sentinels. .. However, Leo will always *read* such sentinels correctly. ``@+all`` and ``-all`` These sentinels mark the range of``@all`` directives. ``@at`` and ``@doc`` These sentinels mark the range of ``@doc`` parts. .. Leo 5.7.4. retired the **trailing whitespace convention**: .. A line in a doc part is followed by an inserted newline .. if and only if the newline if preceded by whitespace. .. To make this convention work, Leo's old write code deleted the trailing .. whitespace of all lines that are followed by a "real" newline. .. ``@+body`` ``and @-body`` **(Leo 3.x only)** .. These sentinels marked the range of body text. ``@@delims`` This sentinel marks the range of an ``@delims`` directive. |br| The range continues until the next ``@delims`` directive. Adding, deleting or changing ``@@delim`` *sentinels* will destroy Leo's ability to read the external file. Mistakes in using the ``@delims`` *directives* have no effect on Leo, though such mistakes will thoroughly mess up a external file as far as compilers, HTML renderers, etc. are concerned. ``@@section-delims`` This sentinel represents the ``@section-delims`` directive. |br| Such must appear in the root ``@`` node. ``@+leo`` This sentinel marks the start of any external file. |br| This sentinel has the form:: @leo The read code uses single-line comments if ```` is empty. |br| The write code generates single-line comments if possible. The ``@+leo`` sentinel contains other information. For example:: @leo-ver=4-thin ``@-leo`` This sentinel marks the end of the Leo file. |br| Nothing but whitespace should follow this directive. .. @+middle and @-middle: Created in Leo 4.0, removed in Leo 5.3. .. These sentinels marked a range of nodes. .. These sentinels were a **mistake**. See: .. https://github.com/leo-editor/leo-editor/issues/132 ``@+node`` This sentinel marks the start of a node:: @+node:gnx: Nodes continue until the next ``@+node``, ``at-all`` or ``at-others`` sentinel. .. ``@nl`` and ``@nonl``: Leo 3.x only .. These sentinels inserted or suppressed newlines in the outline. .. Modern versions of Leo do not write ``@nonl`` or ``@nl``. .. As a result, body text always ends with at least one newline. ``@+others`` and ``@-others`` These sentinels mark the range of an ``@others`` directive. ``@verbatim`` This sentinel indicates that the next line of the external file is not a sentinel. This escape convention allows body text to contain lines that would otherwise be considered sentinel lines. .. @@verbatimAfterRef .. Old versions of Leo generated @verbatimAfterRef when a comment .. following a section reference would otherwise be treated as a sentinel. The Leonine way to refactor code ++++++++++++++++++++++++++++++++ This paper explains how to use cff (clone-find-flattened) while refactoring code. I could not have completed the refactoring of Leo's atFile write code without using continuous, extensive use of cff. There are two key ideas: 1. The clones produced by cff are short-term or medium-term data, easily created and easily dispensed with. Such clones are valuable, but not precious. They will eventually be discarded. 2. Unlike tags (or any other kind of non-Leonine data), the clones produced by cff can be reorganized. This is the priceless, unique advantage of clones. You don't understand clones if you don't get this. Example 1. While refactoring, it is essential to see all actual uses of a symbol (method, or ivar, whatever). The starting point is to use cff to find all potential uses of the symbol. If multiple files or classes use the symbol, you can use the suboutline-only option to limit the matches created by cff. 2. After finding all potential uses of the symbol, you can reorganize the resulting clones as follows: - Delete nodes that are completely irrelevant. - Squirrel away likely-irrelevant nodes in a new organizer node. - Highlight the defining node, say by making it the preceding sibling of the cff node. - Leave all actual uses of the symbol where they are. 3. You have now focused your attention on the nodes that will likely change. You can now rerun the search only on those cloned nodes to see all instances of the symbol that might be changed. This is a crucial double check on proposed changes. Summary I highly recommend that all Leonine programmers use the approach just described when refactoring code. Neither tags, nor filters, nor refactoring packages can emulate the Leonine way of refactoring. Leo's Colorizer: Theory of Operation ++++++++++++++++++++++++++++++++++++ .. _`QSyntaxHighlighter`: https://doc.qt.io/qt-6/qsyntaxhighlighter.html Are you sure you want to read this appendix? It is intended *only* for Leo's core developers. This appendix contains the theory of operation for the ``JEditColorizer`` (**jedit**) class in ``leoColorizer.py``. Unless qualified, *all methods are members of the jedit class*. **Executive summary** Use the ``--trace=coloring`` to see the colorizer in action. The jedit class collaborates with the **qsh**, a singleton instance of the `QSyntaxHighlighter`_ class. This collaboration allows the qsh to *minimize* the calls to ``jedit.redraw``. The qsh calls ``redraw(s)`` *only* if line ``s`` could possibly need recoloring. Leo's core *must never* call ``redraw``! These calls must happen automatically. Calling ``redraw`` from Leo's core could cause hard-to-find bugs! **States** allow the jedit class to handle constructs like strings or comments that continue from one line to another. **Mode files**, python files in the ``leo/modes`` directory, tell the jedit class how to colorize specific languages. **Pattern matchers** do the actual syntax coloring. Components of the jedit class ***************************** The ``JEditColorizer`` class consists of **drivers**, **pattern matchers** and various **helper utilities**. **Drivers** The ``redraw`` and ``mainLoop`` methods are the interface between the qsh and the methods of the jedit class. ``qsh.rehighlight`` calls ``redraw(s)`` to colorize a *single* line ``s``. After initing the **initial state**, ``redraw`` calls ``mainLoop``. Leo's core *must never* call ``redraw``! Doing so could cause hard-to-find bugs. ``redraw(s)`` calls ``mainLoop(s)`` to do the actual syntax coloring. ``mainLoop(s)`` calls **rule functions** (defined in mode files). Rule functions delegate their work to exactly one **pattern matcher** (described next). In other words, *pattern matchers are callbacks.* **Pattern matchers** Pattern matchers form the bulk of the jedit class. Each pattern matcher does the following: - Examines the string ``s[i:]``, where s is the argument to ``mainLoop(s)``. - Updates the **ending state** for successful matches. - Returns an integer **return code**:: -1: complete failure: mainLoop returns 0: partial failure: mainLoop tries to match other rules >0: success: mainLoop increments i by n and tries to match other rules The big challenge ***************** The qsh can (and will) call ``recolor(s)`` where s is *out-of-sequence* with the preceding call. For example, the following snippet is typical in ``@language jupytext`` nodes: .. code-block:: python # %% (like @language python) def spam(): # %% [markdown] (like @language md) # Section 1 It was the best of times; it was the worst of times. The user can click (or cut or paste) anywhere in this text. The qsh will then call ``recolor(s)`` starting with the first changed line. *There is no necessary relationship between* ``s`` *and the line that* ``recolor`` *last saw*! About states and restarters *************************** States allow the colorizer to handle out-of-order calls to ``recolor``. The colorizer collaborates the qsh in *only one way*, namely by calling ``self.highlighter.setCurrentBlockState(n)``, where ``n`` is an *integer* state number. Internally, the colorizer uses *string* states. The colorizer's **state utilities** handle the conversion between integer and string representations. States represent, for a *particular* line: - Which ``@language`` directive is in effect. - Which coloring directive (``@killcolor``, ``@color``, ``@nocolor``, etc.) is in effect. - Whether the coloring for the line *continues* to the next line. For example, Python docstrings may span multiple lines. A fine point: The state utilities ensure that all integer state numbers are unique and refer to the proper language in effect. For example, if a node contains multiple ``@language`` directives, the default states of the two languages must have different state numbers! The colorizer keeps track of states as follows: ``recolor`` first computes the line's **initial state** from the **ending state** of the *previous* line. By default, the initial state becomes the ending state of *current* state. Pattern matchers usually leave the ending state unchanged. If a pattern matcher matches an ``@language`` or coloring directive, the matcher creates (indirectly, via the state utilities) a new state representing the directive. This is the easy case. If a pattern matcher matches syntax that continues to the next line, the pattern matcher creates a **restarter matcher** by calling ``setRestart``. The restarter matchers are essentially closures, bindings of the pattern matcher *and all of its kwargs*. The details are hairy, but the coding pattern is the same for all cases. See the source code for details. Mode files ********** .. _jEdit: https://www.jedit.org/ The `jEdit`_ editor drives its syntax colorer using ``.xml`` files. Leo uses ``.py`` files instead. Long ago, Leo's ``jEdit2Py`` script created the python **mode files** in the ``leo/modes`` directory. Roughly speaking, there is one mode file for every language that Leo can colorize. Over the years, Leo's devs have modified several of these mode files by hand. ``init`` sets ``jedit.rulesDict`` from the mode file for the language in effect. Keys are **lead-in characters**; values are lists of **rule functions** (**rules** for short). Each mode file defines its own rulesDict and rules. ``mainLoop(s)`` uses the rulesDict to retrieve a list of zero or more rules to be used when scanning s. It's that simple. Legacy mode files could collaborate using the ``delegate`` keyword, but that scheme was (and is) clumsy. Leo 6.8.3 added ``leo/modes/jupytext.py``, a new mode file for ``@language jupytext``. This mode file introduced a new style of collaboration. The mode files for the jupytext, python and markdown languages collaborate *directly* using modifications of *existing* rules in ``python.py`` and ``md.py``. leoAst.py +++++++++ The classes in `leoAst.py `_ unify python's token-based and ast-based worlds by creating two-way links between tokens in the token list and ast nodes in the parse tree. Are you sure you want to use leoAst? - `asttokens `_ will usually be a better choice: - Support for the leoAst module ends with Python 3.14. leoAst.py is part of `Leo `_, and can be used completely independently of Leo. You must use Leo to see the intended outline structure of the code. Without Leo, you will see special **sentinel comments** that create Leo's outline structure. These comments have the form:: #@:..: If you have any trouble installing or using this code, please help for help on `Leo's forum `_ Running leoAst.py ***************** Running leoAst.py from the command line ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ leoAst.py is designed to be run from the command line:: usage: leoAst.py --help leoAst.py [--fstringify | --fstringify-diff | --orange | --orange-diff] PATHS leoAst.py --py-cov [ARGS] leoAst.py --pytest [ARGS] leoAst.py --unittest [ARGS] examples: --py-cov "-f TestOrange" --pytest "-f TestOrange" --unittest TestOrange positional arguments: PATHS directory or list of files optional arguments: -h, --help show this help message and exit --fstringify leonine fstringify --fstringify-diff show fstringify diff --orange leonine Black --orange-diff show orange diff --py-cov run pytest --cov on leoAst.py --pytest run pytest on leoAst.py --unittest run unittest on leoAst.py Running the code python programmatically ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ To access the code, do one of the following: .. code-block:: import leoAst import leo.core.leoAst as leoAst You can then run the fstringify commands as follows: .. code-block:: changed = leoAst.Fstringify().fstringify_file(filename) changed = leoAst.Fstringify().fstringify_diff_files(filename) Running unit tests and coverage tests programmatically ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ The following runs all unit tests for leoAst.py:: python -m leo.core.leoAst The following runs coverage tests:: pytest -x --cov-report html --cov-report term-missing --cov=leo.core.leoAst leo/core/leoAst.py TokenOrder classes: Theory of operation *************************************** This is the Theory of Operation for the TokenOrderGenerator (TOG) class and related classes. Token-order classes ^^^^^^^^^^^^^^^^^^^ The **TokenOrderGenerator** (TOG) class injects two-way links between all tokens and the corresponding ast nodes. The TOG class also injects parent/child links into all ast nodes. The TOG class defines visitors that visit ast nodes in **token order**, the traversal order that corresponds to the order of the tokens produced by python's tokenizer module. The only way to ensure this correspondence is to use separate visitors for all ast nodes. All visitors are straightforward generators. TOG visitors eventually call ``TOG.sync_token``, which checks that that tokens are, in fact, visited in the correct order. ``TOG.sync_token`` is an ever-present unit test. The **TokenOrderTraversal** (TOT) class uses the parent/child links created by the TOG class. TOT.traverse contains a single for-loop that calls all nodes of the parse tree in token order. This loop is extremely fast. Using the TOT class, client code can easily modify the token list or parse tree as desired. Other classes ^^^^^^^^^^^^^ The **Token** class represents one token, created by tokenize.tokenize. The **Fstringify** class is an re-implementation of the external fstringify project using the TOG class. The **Orange** class is a re-implementation of the black project. |br| The name "Orange" is a play on words: "Orange is the new black". The **AstDumper** class provides an extensive set of tools for examining token lists, parse trees, and the links between them. The **BaseTest** class provides common infrastructure for all other test classes. *Important*: BaseTest.make_data is, all by itself, a very strong unit test. Significant vs insignificant tokens ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ The distinction between **significant** and **insignificant** tokens is crucial. Visitors call ``TOG.gen_token``, ``TOG.gen_op``, etc. *only* for significant tokens. The ``is_significant`` and ``is_significant_token`` functions define which tokens are significant. Visitors can't know, just by looking at the parse tree, whether the input contains *insignificant* tokens. For example, the source tokens might contain non-essential parentheses, or optional trailing commas, or whether two statements are separated by a semicolon. Helping TOG visitors ^^^^^^^^^^^^^^^^^^^^ The ``TOG.do_If`` visitor calls ``TOG.find_next_significant_token`` to determine whether ``TOG.do_If`` should generate an "if" or an "elif" token. This help is essential, because the following two source files generate identical parse trees! :: if 1: if 1: pass pass else: elif 2: if 2: pass pass Similarly, the ``TOG.do_Str`` and ``TOG.do_JoinedStr`` visitors call ``TOG.get_concatenated_string_tokens`` to handle one ore more concatenated string tokens. Finally, ``TOG.do_slice`` calls ``TOG.find_next_significant_token`` to determine whether a slice without a step contains an optional colon. ``TOG.find_next_significant_token`` and ``TOG.get_concatenated_string_tokens`` are crucial inventions. TOG class would not be possible without them. Syncing tokens ^^^^^^^^^^^^^^ ``TOG.px`` is an index into the token list. It is either -1, or it points at the previous significant token. *Note*: TOG.find_next_significant_token and TOG.get_concatenated_string_tokens use TOG.px, but never change TOG.px. TOG.sync_token(self, kind, val) associates tokens with ast nodes as follows: 1. If (kind, val) denote an *insignificant* token, TOG.sync_token does nothing. 2. Otherwise, (kind, val) denotes a *significant* token. TOG.sync_token associates that token, *plus* all previous *insignificant* tokens with self.node, the ast node presently being visited. In addition, if (kind, val) denotes a *significant* token, TOG.sync_token checks that the next *significant* token in the token list has the expected kind and value. This is done as follows: - TOG.sync_token advances TOG.px to point at the next significant token, call it T. - TOG.raises AssignLinksError if (T.kind, T.value) != (kind, val) To summarize token syncing: - The TOG.px index tracks the last-seen *significant* token. - TOG.px advances monotonically through the token list. - TOG.find_next_significant_token and TOG.get_concatenated_string_tokens scan forward through the token list using a private copy of TOG.px. These methods never change TOG.px itself. - This token-syncing machinery is the *simplest* thing that could possibly work. It is also the *fastest* thing that could possibly work. Figures of merit ^^^^^^^^^^^^^^^^ **Simplicity**: - The distinction between significant and insignificant tokens makes token-order traversals possible. This distinction drastically simplifies TOG visitors. They never have to generate insignificant tokens! - TOG.find_next_significant_token and TOG.get_concatenated_string_tokens() use TOG.px to look ahead in the token list. - It took a long time to realize that the parse tree needs help from the token list, not the other way around! **Speed**: The TOG creates links between tokens and ast nodes in roughly the time taken by python's tokenize.tokenize and ast.parse library methods. The TOT class traverses trees annotated with parent/child links even more quickly. TOG class avoids both ``ast.fix_missing_locations`` and ``ast.get_source_segment``, which are too slow to be useful. **Memory**: The TOG class makes no significant demand on python's resources: - Generators add nothing to python's call stack. - The *only* variable-length data created by the TOG is TOG.node_stack. This stack resides in python's heap, so its length is unimportant. In the worst case, it might contain a few thousand entries. - The TOT uses no variable-length data whatever. Maintaining leoAst.py ********************* .. _`Leo issue #1440`: https://github.com/leo-editor/leo-editor/issues/1440 New ast nodes are sometimes required to support new language features, especially language features that require new syntax or keywords. Python has added new nodes fairly often in the past. New nodes may be added in future. When that happens, the following changes will be needed to leoAst.py: - Add a visitor for the new node. - Add one or more unit tests that fully cover the new visitor. The ``test_visitors_exist`` unit test checks that visitors exist for all ast nodes defined by to a particular version of python. See `Leo issue #1440`_ for notes relating to the code. Unicode reference +++++++++++++++++ Leo uses unicode internally for all strings. 1. Leo converts headline and body text to unicode when reading .leo files and external files. Both .leo files and external files may specify their encoding. The default is utf-8. If the encoding used in a external file is not "utf-8" it is represented in the @+leo sentinel line. For example:: #@+leo-encoding=iso-8859-1. The utf-8 encoding is a "lossless" encoding (it can represent all unicode code points), so converting to and from utf-8 plain strings will never cause a problem. When reading or writing a character not in a "lossy" encoding, Leo converts such characters to '?' and issues a warning. 2. When writing .leo files and external files Leo uses the same encoding used to read the file, again with utf-8 used as a default. 3. leoSettings.leo contains the following Unicode settings, with the defaults as shown:: default_derived_file_encoding = UTF-8 new_leo_file_encoding = UTF-8 These control the default encodings used when writing external files and .leo files. Changing the new_leo_file_encoding setting is not recommended. See the comments in leoSettings.leo. You may set default_derived_file_encoding to anything that makes sense for you. 4. The @encoding directive specifies the encoding used in a external file. You can't mix encodings in a single external file. Valid URL's +++++++++++ Leo checks that the URL is valid before attempting to open it. A valid URL is: - 3 or more lowercase alphas - followed by one : - followed by one or more of: - ``$%&'()*+,-./0-9:=?@A-Z_a-z{}~`` - followed by one of: ``$%&'()*+/0-9:=?@A-Z_a-z}~`` That is, a comma, hyphen and open curly brace may not be the last character. URL's in Leo should contain no spaces: use %20 to indicate spaces. You may use any type of URL that your browser supports: http, mailto, ftp, file, etc. The Mulder/Ream update algorithm ++++++++++++++++++++++++++++++++ This appendix documents the Mulder/Ream update algorithm in detail, with an informal proof of its correctness. Prior to Leo 5.1, Leo used Bernhard Mulder's original algorithm to read @shadow files. Starting with Leo 5.1, Leo uses this algorithm to read both @clean and @shadow files. Conceptually, both algorithms work as described in the next section. In February 2015 EKR realized that the @shadow algorithm could be used to update @clean (@nosent) files. Simplifying the algorithm instantly became a top priority. The new code emerged several days later, made possible by the x.sentinels array. It is an important milestone in Leo's history. What the algorithm does *********************** For simplicity, this discussion will assume that we are updating an external file, x, created with @clean x. The update algorithm works exactly the same way with @shadow trees. The algorithm works with *any* kind of text file. The algorithm uses only difflib. It knows nothing about the text or its meaning. No parsing is ever done. Suppose file x has been changed outside of Leo. When Leo reads x it does the following: 1. Recreates the *old* version of x *without* sentinels by writing the @clean x *outline* into a string, as if it were writing the @clean x outline again. 2. Recreates all the lines of x *with* sentinels by writing the @clean x *outline* into a string, as if it was writing an @file node! Let's call these lines the **old sentinels** lines. 3. Uses difflib.SequenceMatcher to create a set of diffs between the old and new versions of x *without* sentinels. **Terminology**: the diffs tell how to change file a into file b. The actual code uses this terminology: **a** is set of lines in the old version of x, **b** is the set of lines in the new version of x. 4. Creates a set of lines, the **new sentinels lines** using the old sentinels lines, the a and b lines and the diffs. This is the magic. Bernhard Mulder's genius was conceiving that a three-way merge of lines could produce the new outline, *with* sentinels. The code is in x.propagate_changed_lines and its helpers. 5. Replaces the @clean tree with the new tree created by reading the new sentinels lines with the @file read logic. **Important**: The update algorithm never changes sentinels. It never inserts or deletes nodes. The user is responsible for creating nodes to hold new lines, or for deleting nodes that become empty as the result of deleting lines. Guesses don't matter ******************** There are several boundary cases that the update algorithm can not resolve. For example, if a line is inserted between nodes, the algorithm can not determine whether the line should be inserted at the end of one node or the start of the next node. Let us call such lines **ambiguous lines**. The algorithm *guesses* that ambiguous lines belongs at the end of a node rather than at the start of the next node. This is usually what is wanted--we usually insert lines at the end of a node. Happily, **guesses don't matter**, for the following reasons: 1. The external file that results from writing the @clean x tree will be the same as the updated external file *no matter where* ambiguous lines are placed. In other words, the update algorithm is **sound**. 2. Leo reports nodes that were changed when reading any external file. The user can review changes to @clean and @file trees in the same way. 3. The user can permanently correct any mistaken guess. Guesses only happen for *newly inserted or changed* lines. Moving an ambiguous line to the following node will not change the external file. As a result, the next time Leo reads the file the line will be placed in the correct node! This proves that @shadow and @clean are easy and safe to use. The remaining sections of this document discuss code-level details. Background of the code ********************** The algorithm depends on three simple, guaranteed, properties of SequenceMatcher.opcodes. See https://docs.python.org/2/library/difflib.html#sequencematcher-examples **Fact 1**: The opcodes tell how to turn x.a (a list of lines) into x.b (another list of lines). The code uses the a and b terminology. It's concise and easy to remember. **Fact 2**: The opcode indices ai, aj, bi, bj *never* change because neither x.a nor x.b changes. Plain lines of the result can be built up by copying lines from x.b to x.results:: 'replace' x.results.extend(x.b[b1:b2]) 'delete' do nothing (b1 == b2) 'insert' x.results.extend(x.b[b1:b2]) 'equal' x.results.extend(x.b[b1:b2]) **Fact 3**: The opcodes *cover* both x.a and x.b, in order, without any gaps. This is an explicit requirement of sm.get_opcode: - The first tuple has ai==aj==bi==bj==0. - Remaining tuples have ai == (aj from the preceding tuple) and bi == (bj from the previous tuple). Keep in mind this crucial picture: - The slices x.a[ai:aj] cover the x.a array, in order without gaps. - The slices x.b[bi:bj] cover the x.b array, in order without gaps. Aha: the x.sentinels array ************************** Mulder's original algorithm was hard to understand or to change. The culprit was the x.mapping array, which mapped indices into arrays of lines *with* sentinels to indices into arrays of lines *without* sentinels. The new algorithm replaces the x.mapping array with the x.sentinels array. As a result, diff indices never need to be adjusted and handling diff opcodes is easy. For any index i, x.sentinels[i] is the (possibly empty) list of sentinel lines that precede line a[i]. Computing x.sentinels from old_private_lines is easy. Crucially, x.a and x.sentinels are *parallel arrays*. That is, len(x.a) == len(x.sentinels), so indices into x.a are *also* indices into x.sentinels. Strategy & proof of correctness ******************************* Given the x.sentinels array, the strategy for creating the results is simple. Given indices ai, aj, bi, bj from an opcode, the algorithm: - Writes sentinels from x.sentinels[i], for i in range(ai,aj). - Writes plain lines from b[i], for i in range(bi,bj). This "just works" because the indices cover both a and b. - The algorithm writes sentinels exactly once (in order) because each sentinel appears in x.sentinels[i] for some i in range(len(x.a)). - The algorithm writes plain lines exactly once (in order) because each plain line appears in x.b[i] for some i in range(len(x.b)). This completes an informal proof of the correctness of the algorithm. The leading and trailing sentinels lines are easy special cases. This code, appearing before the main loop, ensures that leading lines are written first, and only once: .. code-block:: x.put_sentinels(0) x.sentinels[0] = [] Similarly, this line, at the end of the main loop, writes trailing sentinels: .. code-block:: x.results.extend(x.trailing_sentinels) Summary ******* The algorithm creates an updated set of lines *with* sentinels using the @clean outline and the updated external file. These new lines then replace the original @clean with a new @clean tree. The algorithm uses only difflib. It will work with *any* kind of text file. No knowledge of any language is needed. The algorithm depends on simple, guaranteed, properties of indices in SequenceMatcher opcodes. The algorithm steps through x.sentinels and x.b, extending x.results as it goes. The algorithm gets all needed data directly from opcode indices into x.sentinels and x.b. Using opcode indices requires neither reader classes nor auxiliary indices. The algorithm is simple enough to be understood at first reading. I'll remember its details for the rest of my life. Why I like Python +++++++++++++++++ I wrote this soon after discovering Python in 2001. The conclusions are still valid today. I've known for a while that Python was interesting; I attended a Python conference last year and added Python support to Leo. But last week I got that Python is something truly remarkable. I wanted to convert Leo from wxWindows to wxPython, so I began work on c2py, a Python script that would help convert from C++ syntax to Python. While doing so, I had an Aha experience. Python is more than an incremental improvement over Smalltalk or C++ or objective-C; it is "something completely different". The rest of this post tries to explain this difference. Clarity ******* What struck me first as I converted C++ code to Python is how much less blah, blah, blah there is in Python. No braces, no stupid semicolons and most importantly, *no declarations*. No more pointless distinctions between const, char \*, char const \*, char \* and wxString. No more wondering whether a variable should be signed, unsigned, short or long. Declarations add clutter, declarations are never obviously right and declarations don't prevent memory allocation tragedies. Declarations also hinder prototyping. In C++, if I change the type of something I must change all related declarations; this can be a huge and dangerous task. With Python, I can change the type of an object without changing the code at all! It's no accident that Leo's new log pane was created first in Python. Functions returning tuples are a "minor" feature with a huge impact on code clarity. No more passing pointers to data, no more defining (and allocating and deallocating) temporary structs to hold multiple values. .. _`pylint`: https://pylint.pycqa.org/en/latest/ Python can't check declarations because there aren't any. However, there is a really nifty tool called `pylint`_ that does many of the checks typically done by compilers. Power ***** Python is much more powerful than C++, not because Python has more features, but because Python needs *less* features. Some examples: - Python does everything that the C++ Standard Template Library (STL) does, without any of the blah, blah, blah needed by STL. No fuss, no muss, no code bloat. - Python's slicing mechanism is very powerful and applies to any sequence (string, list or tuple). Python's string library does more with far less functions because slices replace many functions typically found in other string libraries. - Writing dict = {} creates a dictionary (hash table). Hash tables can contain anything, including lists and other hash tables. - Python's special functions, __init__, __del__, __repr__, __cmp__, etc. are an elegant way to handle any special need that might arise. Safety ****** Before using Python I never fully realized how difficult and dangerous memory allocation is in C++. Try doing: .. code-block:: aList[i:j] = list(aString) in C. You will write about 20 lines of C code. Any error in this code will create a memory allocation crash or leak. Python is fundamentally safe. C++ is fundamentally unsafe. When I am using Python I am free from worry and anxiety. When I am using C++ I must be constantly "on guard." A momentary lapse can create a hard-to-find pointer bug. With Python, almost nothing serious can ever go wrong, so I can work late at night, or after a beer. The Python debugger is always available. If an exception occurs, the debugger/interpreter tells me just what went wrong. I don't have to plan a debugging strategy! Finally, Python recovers from exceptions, so Leo can keep right on going even after a crash! Speed ***** Python has almost all the speed of C. Other interpretive environments such as icon and Smalltalk have clarity, power and safety similar to Python. What makes Python unique is its seamless way of making C code look like Python code. Python executes at essentially the speed of C code because most Python modules are written in C. The overhead in calling such modules is negligible. Moreover, if code is too slow, one can always create a C module to do the job. In fact, Python encourages optimization by moving to higher levels of expression. For example, Leo's Open command reads an XML file. If this command is too slow I can use Python's XML parser module. This will speed up Leo while at the same time raising the level of the code. Conclusions *********** Little of Python is completely new. What stands out is the superb engineering judgment evident in Python's design. Python is extremely powerful, yet small, simple and elegant. Python allows me to express my intentions clearly and at the highest possible level. The only hope of making Leo all it can be is to use the best possible tools. I believe Python will allow me to add, at long last, the new features that Leo should have. Edward K. Ream, October 25, 2001. P.S., September, 2005: Four years of experience have only added to my admiration for Python. Leo could not possibly be what it is today without Python.