splc 0.1.1

splc 0.1.1

A small release on top of 0.1.0: the compiler builds and passes its tests on every
LLVM from 18 to 23, and the verse generator now writes iambic pentameter that reads as
such, not only one that passes the checker.

Changes

• tools/text_to_spl.py --verse plans a speech syllable by syllable from the dictionary
  stress of every word (function words weak, nouns strong, adjectives and two-syllable
  nouns trochaic), keeps the stress strictly alternating through the whole speech and
  never splits a word across a line. Powers of two are spelled with trochaic adjectives
  (a lovely golden cat) and larger values with twice, the square of, the cube of
  and the sum of as the position in the line allows. The old mode relied on the
  checker's rule that a monosyllable may fall in either position, which let lines like
  a big big big big cat through.
• examples/balcony_verse.spl is regenerated: 2,166 lines, all 1,083 couplets rhymed,
  passes -fpentameter=error -frhyme-scheme=AABB, output unchanged.
• Code generation no longer relies on BasicBlock::getTerminator() to detect an empty
  block, which on LLVM 23 left the blocks after a goto inside an If unterminated and
  failed module verification. The LLVM 21 and later target API changes (Triple-typed
  lookups) are handled as well.
• Release notes and README refreshed; docs/releasing.md describes how a release is cut.

Building

CMake 3.20 or newer, a C++17 compiler and LLVM 18 or newer. CI runs the seventeen tests
and a calibration check on Ubuntu (apt LLVM 18) and macOS (Homebrew LLVM, currently 23).

Licence

GPL-3.0-or-later; see LICENSE and THIRD_PARTY_LICENSES.md.

v0.1.0

splc 0.1.0

First release of splc, an LLVM front end for the Shakespeare Programming Language that
compiles plays to native code and can insist that they be written in verse.

What it is

The Shakespeare Programming Language (Hasselström and Wiberg, 2001) is an esoteric
language in which programs are plays: characters are variables, dialogue is code, and
"Thou art the sum of thyself and a cat" is an increment. The original implementation
recognises a few hundred hand-picked words. splc recognises 171,534, generated from
CMUdict, WordNet and the VADER sentiment lexicon, so that any English noun or adjective
works and any word's polarity (a cat is 1, a pig is -1) comes from data rather than a
list. It emits LLVM IR, runs the standard optimisation pipeline, and links a small C
runtime that tracks who is on stage.

Because the lexicon also carries stress and pronunciation, the compiler can check the
poetry. -fpentameter scans every line of dialogue as iambic pentameter, with the
Elizabethan latitude a real editor allows: promoted syllables, inverted first feet and
feet after punctuation, feminine endings, syncope, elision across words, the period stress
of words like aspect and revenue, and the metrical stress of 596 Shakespearean names.
-fcouplets, -frhyme-scheme and -fnear-rhymes check rhyme, and -fsonnet requires
every speech to be a sonnet. examples/sonnet.spl is a program of two sonnets that prints
Hi! and Ho! and passes -fsonnet -fpentameter=error. examples/balcony.spl prints Juliet's
balcony soliloquy in prose, and examples/balcony_verse.spl, generated by
tools/text_to_spl.py --verse, prints the same text in rhymed iambic pentameter and passes
-fpentameter=error -frhyme-scheme=AABB.

Calibrated on the complete works

The checkers were tuned and measured against Shakespeare himself. The repository carries,
as plain text extracted from the shakespeare.mit.edu edition, all 154 sonnets, the verse
and prose speeches of all 37 plays (60,292 and 16,744 lines), and the three long poems as
stanzas.

At the default tolerance, 93% of sonnet lines and 86% of the plays' verse lines scan,
against 34% of the prose lines and 59% of sonnet lines with their words shuffled. The
rhyme checker recognises 93% of the sonnets' 1,078 rhyme pairs strictly and 98% with
-fnear-rhymes, with the same figures on Venus and Adonis and Lucrece, which it was not
tuned on.

Two results fell out of the calibration that are worth a look on their own. Ranking the
plays by how much of their verse scans reproduces the accepted chronology: the early
histories, Romeo and Juliet and King John at the top, The Tempest, Coriolanus and
Pericles at the bottom. And the share of lines with a feminine ending, which the checker
counts as a by-product, reproduces Spedding's 1850 test: 7 to 11% in the early plays, 29
to 32% in the late romances, and 44% in Henry VIII, the number Spedding used to argue that
half of that play is Fletcher's.

Tools

• --scan prints the scansion of every line of a play; --scan-text scans any text file;
  --suggest-stress and tools/mine_stress.py mine a corpus for pronunciations the
  lexicon lacks.
• tools/build_corpus.sh rebuilds the corpora from a shakespeare.mit.edu download, and
  tools/gen_lexicon.py rebuilds the lexicon from data/.
• tools/text_to_spl.py turns a text file into a play that prints it, in prose or, with
  --verse, in rhymed pentameter.
• editors/vscode/ provides syntax highlighting for .spl.
• docs/writing-a-sonnet.md explains how to write a program that is a sonnet.

Building

CMake 3.20 or newer, a C++17 compiler and LLVM 18 or newer. Verified on Linux x86_64 with
LLVM 18.1.3 and 20.1.2 and on macOS (Apple silicon) with Homebrew LLVM 20.1.8 and 23.1.0;
build_mac.sh does the Homebrew configuration for you. CI runs the seventeen tests and a
calibration check on Ubuntu (apt LLVM 18) and macOS (Homebrew LLVM, currently 23).

The LLVM 21 and later API changes (Triple-typed target lookup) and an LLVM 23 behaviour
change in BasicBlock::getTerminator() on empty blocks are handled in this release.

Licence

GPL-3.0-or-later. The bundled dictionaries are under their own permissive licences
(CMUdict BSD, WordNet Princeton, VADER MIT) and the Shakespeare text is public domain;
see THIRD_PARTY_LICENSES.md.