| 1 | #!/usr/bin/env python2
|
| 2 | """
|
| 3 | Pretty printing library.
|
| 4 |
|
| 5 | Pretty printing means intelligently choosing whitespace including indentation
|
| 6 | and newline placement, to attempt to display data nicely while staying within a
|
| 7 | maximum line width.
|
| 8 | """
|
| 9 |
|
| 10 | # ~~~ Architecture ~~~
|
| 11 | #
|
| 12 | # Based on a version of the algorithm from Wadler's "A Prettier Printer".
|
| 13 | #
|
| 14 | # Pretty printing proceeds in two phases:
|
| 15 | #
|
| 16 | # 1. Convert the thing you want to print into a `doc`.
|
| 17 | # 2. Print the `doc` using a standard algorithm.
|
| 18 | #
|
| 19 | # This separation keeps the details of the data you want to print separate from
|
| 20 | # the printing algorithm.
|
| 21 |
|
| 22 | # ~~~ Pretty Printing Overview ~~~
|
| 23 | #
|
| 24 | # If you're just using this file, you don't need to know how pretty printing
|
| 25 | # works. Just call `PrettyPrinter().PrintValue()`. However if you want to change
|
| 26 | # or extend how values are printed, you'll need to know, so here's an overview.
|
| 27 | #
|
| 28 | # You may want to first read Walder's "A Prettier Printer", which this is based
|
| 29 | # off of:
|
| 30 | # https://homepages.inf.ed.ac.uk/wadler/papers/prettier/prettier.pdf
|
| 31 | #
|
| 32 | # Some additional reading, though only tangentially related:
|
| 33 | #
|
| 34 | # - https://homepages.inf.ed.ac.uk/wadler/papers/prettier/prettier.pdf
|
| 35 | # - https://lindig.github.io/papers/strictly-pretty-2000.pdf
|
| 36 | # - https://justinpombrio.net/2024/02/23/a-twist-on-Wadlers-printer.html
|
| 37 | # - https://lobste.rs/s/1r0aak/twist_on_wadler_s_printer
|
| 38 | # - https://lobste.rs/s/aevptj/why_is_prettier_rock_solid
|
| 39 | #
|
| 40 | # ~ Constructors ~
|
| 41 | #
|
| 42 | # There are just a few constructors for `doc`, from which everything else is
|
| 43 | # built from.
|
| 44 | #
|
| 45 | # Text(string) prints like:
|
| 46 | # |string
|
| 47 | #
|
| 48 | # Break(string) prints like:
|
| 49 | # |string
|
| 50 | # or like a newline:
|
| 51 | # |
|
| 52 | # |
|
| 53 | # (It does the latter if printed in "flat" mode, and the former otherwise. See
|
| 54 | # Group for details.)
|
| 55 | #
|
| 56 | # Concat(a, b) prints like:
|
| 57 | # |AAAAA
|
| 58 | # |AAABBB
|
| 59 | # |BBBBB
|
| 60 | #
|
| 61 | # Indent(3, a) prints like:
|
| 62 | # |AAAAA
|
| 63 | # | AAAAA
|
| 64 | # | AAAAA
|
| 65 | #
|
| 66 | # Group(a) makes a decision. It either:
|
| 67 | # - Prints `a` "flat", meaning that (i) every Break inside of it is printed as a
|
| 68 | # string instead of as a newline, and (ii) every Group nested inside of it is
|
| 69 | # printed flat.
|
| 70 | # - Prints `a` normally, meaning that (i) the Breaks inside of it are printed as
|
| 71 | # newlines, and (ii) the Groups inside of it make their own decision about
|
| 72 | # whether to be flat.
|
| 73 | # It makes this decision greedily. If the current line would not overflow if the
|
| 74 | # group printed flat, then it will print flat. This takes into account not only
|
| 75 | # the group itself, but the content before and after it on the same line.
|
| 76 | #
|
| 77 | # IfFlat(a, b) prints a if in flat mode or b otherwise.
|
| 78 | #
|
| 79 | # Flat(a) prints a in flat mode. You should generally not need to use it.
|
| 80 | #
|
| 81 | # ~ Measures ~
|
| 82 | #
|
| 83 | # The algorithm used here is close to the one originally described by Wadler,
|
| 84 | # but it precomputes a "measure" for each node in the `doc`. This "measure"
|
| 85 | # allows each Groups to decide whether to print flat or not without needing to
|
| 86 | # look ahead per Wadler's algorithm. A measure has two pieces of information:
|
| 87 | #
|
| 88 | # - Measure.flat is the width of the doc if it's printed flat.
|
| 89 | # - Measure.nonflat is the width of the doc until the _earliest possible_
|
| 90 | # newline, or -1 if it doesn't contain a Break.
|
| 91 | #
|
| 92 | # Measures are used in two steps:
|
| 93 | # (1) First, they're computed bottom-up on the `doc`, measuring the size of each
|
| 94 | # node.
|
| 95 | # (2) Later, PrintDoc() stores a measure in each DocFragment. These Measures
|
| 96 | # measure something different: the width from the doc _to the end of the
|
| 97 | # entire doc tree_. This second set of Measures (the ones in the
|
| 98 | # DocFragments) are computed top-down, and they're used to decide for each
|
| 99 | # Group whether to use flat mode or not, without needing to scan ahead.
|
| 100 |
|
| 101 | from __future__ import print_function
|
| 102 |
|
| 103 | from _devbuild.gen.pretty_asdl import (doc, doc_e, DocFragment, Measure,
|
| 104 | MeasuredDoc, List_Measured)
|
| 105 | from mycpp.mylib import log, tagswitch, BufWriter
|
| 106 | from typing import cast, List
|
| 107 |
|
| 108 | _ = log
|
| 109 |
|
| 110 | #
|
| 111 | # Measurements
|
| 112 | #
|
| 113 |
|
| 114 |
|
| 115 | def _EmptyMeasure():
|
| 116 | # type: () -> Measure
|
| 117 | """The measure of an empty doc."""
|
| 118 | return Measure(0, -1)
|
| 119 |
|
| 120 |
|
| 121 | def _FlattenMeasure(measure):
|
| 122 | # type: (Measure) -> Measure
|
| 123 | """The measure if its document is rendered flat."""
|
| 124 | return Measure(measure.flat, -1)
|
| 125 |
|
| 126 |
|
| 127 | def _ConcatMeasure(m1, m2):
|
| 128 | # type: (Measure, Measure) -> Measure
|
| 129 | """Compute the measure of 2 concatenated docs.
|
| 130 |
|
| 131 | If m1 and m2 are the measures of doc1 and doc2, then _ConcatMeasure(m1, m2)
|
| 132 | is the measure of doc.Concat([doc1, doc2]). This concatenation is
|
| 133 | associative but not commutative.
|
| 134 | """
|
| 135 | if m1.nonflat != -1:
|
| 136 | return Measure(m1.flat + m2.flat, m1.nonflat)
|
| 137 | elif m2.nonflat != -1:
|
| 138 | return Measure(m1.flat + m2.flat, m1.flat + m2.nonflat)
|
| 139 | else:
|
| 140 | return Measure(m1.flat + m2.flat, -1)
|
| 141 |
|
| 142 |
|
| 143 | def _SuffixLen(measure):
|
| 144 | # type: (Measure) -> int
|
| 145 | """The width until the earliest possible newline, or end of document."""
|
| 146 | if measure.nonflat != -1:
|
| 147 | return measure.nonflat
|
| 148 | else:
|
| 149 | return measure.flat
|
| 150 |
|
| 151 |
|
| 152 | #
|
| 153 | # Doc Construction
|
| 154 | #
|
| 155 |
|
| 156 |
|
| 157 | def AsciiText(string):
|
| 158 | # type: (str) -> MeasuredDoc
|
| 159 | """Print string (which must not contain a newline)."""
|
| 160 | return MeasuredDoc(doc.Text(string), Measure(len(string), -1))
|
| 161 |
|
| 162 |
|
| 163 | def _Break(string):
|
| 164 | # type: (str) -> MeasuredDoc
|
| 165 | """If in flat mode, print string, otherwise print `\n`.
|
| 166 |
|
| 167 | Note: Doesn't try to compute Unicode width, since we control these strings.
|
| 168 | """
|
| 169 | return MeasuredDoc(doc.Break(string), Measure(len(string), 0))
|
| 170 |
|
| 171 |
|
| 172 | def _Indent(indent, mdoc):
|
| 173 | # type: (int, MeasuredDoc) -> MeasuredDoc
|
| 174 | """Add 'indent' spaces after every newline in mdoc."""
|
| 175 | return MeasuredDoc(doc.Indent(indent, mdoc), mdoc.measure)
|
| 176 |
|
| 177 |
|
| 178 | def _Splice(out, mdocs):
|
| 179 | # type: (List[MeasuredDoc], List[MeasuredDoc]) -> Measure
|
| 180 | """Optimization for _Concat.
|
| 181 |
|
| 182 | This reduces the total size of the doc_t tree, and thus the memory usage
|
| 183 | (as long as we mylib.MaybeCollect() in _HNode!)
|
| 184 |
|
| 185 | Example of optimizing _Concat nodes together: _Field() concatenates
|
| 186 | AsciiText("name:") and _HNode(), and the latter is often a doc.Concat node.
|
| 187 | """
|
| 188 | measure = _EmptyMeasure()
|
| 189 | for mdoc in mdocs:
|
| 190 | with tagswitch(mdoc.doc) as case:
|
| 191 | if case(doc_e.Concat):
|
| 192 | child = cast(List_Measured, mdoc.doc)
|
| 193 | # ignore return value, because the parent has the measure already
|
| 194 | _Splice(out, child)
|
| 195 | else:
|
| 196 | out.append(mdoc)
|
| 197 | measure = _ConcatMeasure(measure, mdoc.measure)
|
| 198 | return measure
|
| 199 |
|
| 200 |
|
| 201 | def _Concat(mdocs):
|
| 202 | # type: (List[MeasuredDoc]) -> MeasuredDoc
|
| 203 | """Print the mdocs in order (with no space in between)."""
|
| 204 | flattened = List_Measured.New()
|
| 205 | measure = _Splice(flattened, mdocs)
|
| 206 | return MeasuredDoc(flattened, measure)
|
| 207 |
|
| 208 |
|
| 209 | def _Group(mdoc):
|
| 210 | # type: (MeasuredDoc) -> MeasuredDoc
|
| 211 | """Print mdoc. Use flat mode if mdoc will fit on the current line."""
|
| 212 | return MeasuredDoc(mdoc, mdoc.measure)
|
| 213 |
|
| 214 |
|
| 215 | def _IfFlat(flat_mdoc, nonflat_mdoc):
|
| 216 | # type: (MeasuredDoc, MeasuredDoc) -> MeasuredDoc
|
| 217 | """If in flat mode, print flat_mdoc; otherwise print nonflat_mdoc."""
|
| 218 | return MeasuredDoc(
|
| 219 | doc.IfFlat(flat_mdoc, nonflat_mdoc),
|
| 220 | Measure(flat_mdoc.measure.flat, nonflat_mdoc.measure.nonflat))
|
| 221 |
|
| 222 |
|
| 223 | def _Flat(mdoc):
|
| 224 | # type: (MeasuredDoc) -> MeasuredDoc
|
| 225 | """Prints mdoc in flat mode."""
|
| 226 | return MeasuredDoc(doc.Flat(mdoc), _FlattenMeasure(mdoc.measure))
|
| 227 |
|
| 228 |
|
| 229 | class PrettyPrinter(object):
|
| 230 |
|
| 231 | def __init__(self, max_width):
|
| 232 | # type: (int) -> None
|
| 233 | self.max_width = max_width
|
| 234 |
|
| 235 | def _Fits(self, prefix_len, group, suffix_measure):
|
| 236 | # type: (int, MeasuredDoc, Measure) -> bool
|
| 237 | """Will group fit flat on the current line?"""
|
| 238 | measure = _ConcatMeasure(_FlattenMeasure(group.measure),
|
| 239 | suffix_measure)
|
| 240 | return prefix_len + _SuffixLen(measure) <= self.max_width
|
| 241 |
|
| 242 | def PrintDoc(self, document, buf):
|
| 243 | # type: (MeasuredDoc, BufWriter) -> None
|
| 244 | """Pretty print a doc_t to a BufWriter."""
|
| 245 |
|
| 246 | # The width of the text we've printed so far on the current line
|
| 247 | prefix_len = 0
|
| 248 | # A _stack_ of document fragments to print. Each fragment contains:
|
| 249 | # - A MeasuredDoc (doc node and its measure, saying how "big" it is)
|
| 250 | # - The indentation level to print this doc node at.
|
| 251 | # - Is this doc node being printed in flat mode?
|
| 252 | # - The measure _from just after the doc node, to the end of the entire document_.
|
| 253 | # (Call this the suffix_measure)
|
| 254 | fragments = [DocFragment(_Group(document), 0, False, _EmptyMeasure())]
|
| 255 |
|
| 256 | max_stack = len(fragments)
|
| 257 |
|
| 258 | while len(fragments) > 0:
|
| 259 | max_stack = max(max_stack, len(fragments))
|
| 260 |
|
| 261 | frag = fragments.pop()
|
| 262 | UP_doc = frag.mdoc.doc
|
| 263 | with tagswitch(UP_doc) as case:
|
| 264 |
|
| 265 | if case(doc_e.Text):
|
| 266 | text = cast(doc.Text, UP_doc)
|
| 267 | buf.write(text.string)
|
| 268 | prefix_len += frag.mdoc.measure.flat
|
| 269 |
|
| 270 | elif case(doc_e.Break):
|
| 271 | break_ = cast(doc.Break, UP_doc)
|
| 272 | if frag.is_flat:
|
| 273 | buf.write(break_.string)
|
| 274 | prefix_len += frag.mdoc.measure.flat
|
| 275 | else:
|
| 276 | buf.write('\n')
|
| 277 | buf.write_spaces(frag.indent)
|
| 278 | prefix_len = frag.indent
|
| 279 |
|
| 280 | elif case(doc_e.Indent):
|
| 281 | indented = cast(doc.Indent, UP_doc)
|
| 282 | fragments.append(
|
| 283 | DocFragment(indented.mdoc,
|
| 284 | frag.indent + indented.indent,
|
| 285 | frag.is_flat, frag.measure))
|
| 286 |
|
| 287 | elif case(doc_e.Concat):
|
| 288 | concat = cast(List_Measured, UP_doc)
|
| 289 |
|
| 290 | # If we encounter Concat([A, B, C]) with a suffix measure M,
|
| 291 | # we need to push A,B,C onto the stack in reverse order:
|
| 292 | # - C, with suffix_measure = B.measure + A.measure + M
|
| 293 | # - B, with suffix_measure = A.measure + M
|
| 294 | # - A, with suffix_measure = M
|
| 295 | measure = frag.measure
|
| 296 | for mdoc in reversed(concat):
|
| 297 | fragments.append(
|
| 298 | DocFragment(mdoc, frag.indent, frag.is_flat,
|
| 299 | measure))
|
| 300 | # TODO: this algorithm allocates too much!
|
| 301 | measure = _ConcatMeasure(mdoc.measure, measure)
|
| 302 |
|
| 303 | elif case(doc_e.Group):
|
| 304 | # If the group would fit on the current line when printed
|
| 305 | # flat, do so. Otherwise, print it non-flat.
|
| 306 | group = cast(MeasuredDoc, UP_doc)
|
| 307 | is_flat = self._Fits(prefix_len, group, frag.measure)
|
| 308 | fragments.append(
|
| 309 | DocFragment(group, frag.indent, is_flat, frag.measure))
|
| 310 |
|
| 311 | elif case(doc_e.IfFlat):
|
| 312 | if_flat = cast(doc.IfFlat, UP_doc)
|
| 313 | if frag.is_flat:
|
| 314 | subdoc = if_flat.flat_mdoc
|
| 315 | else:
|
| 316 | subdoc = if_flat.nonflat_mdoc
|
| 317 | fragments.append(
|
| 318 | DocFragment(subdoc, frag.indent, frag.is_flat,
|
| 319 | frag.measure))
|
| 320 |
|
| 321 | elif case(doc_e.Flat):
|
| 322 | flat_doc = cast(doc.Flat, UP_doc)
|
| 323 | fragments.append(
|
| 324 | DocFragment(flat_doc.mdoc, frag.indent, True,
|
| 325 | frag.measure))
|
| 326 |
|
| 327 | if 0:
|
| 328 | log('')
|
| 329 | log('___ MAX DocFragment stack: %d', max_stack)
|
| 330 | log('')
|
| 331 |
|
| 332 |
|
| 333 | # vim: sw=4
|