fish-shell/src/fish_indent_common.cpp

476 lines
18 KiB
C++
Raw Normal View History

Port AST to Rust The translation is fairly direct though it adds some duplication, for example there are multiple "match" statements that mimic function overloading. Rust has no overloading, and we cannot have generic methods in the Node trait (due to a Rust limitation, the error is like "cannot be made into an object") so we include the type name in method names. Give clients like "indent_visitor_t" a Rust companion ("IndentVisitor") that takes care of the AST traversal while the AST consumption remains in C++ for now. In future, "IndentVisitor" should absorb the entirety of "indent_visitor_t". This pattern requires that "fish_indent" be exposed includable header to the CXX bridge. Alternatively, we could define FFI wrappers for recursive AST traversal. Rust requires we separate the AST visitors for "mut" and "const" scenarios. Take this opportunity to concretize both visitors: The only client that requires mutable access is the populator. To match the structure of the C++ populator which makes heavy use of function overloading, we need to add a bunch of functions to the trait. Since there is no other mutable visit, this seems acceptable. The "const" visitors never use "will_visit_fields_of()" or "did_visit_fields_of()", so remove them (though this is debatable). Like in the C++ implementation, the AST nodes themselves are largely defined via macros. Union fields like "Statement" and "ArgumentOrRedirection" do currently not use macros but may in future. This commit also introduces a precedent for a type that is defined in one CXX bridge and used in another one - "ParseErrorList". To make this work we need to manually define "ExternType". There is one annoyance with CXX: functions that take explicit lifetime parameters require to be marked as unsafe. This makes little sense because functions that return `&Foo` with implicit lifetime can be misused the same way on the C++ side. One notable change is that we cannot directly port "find_block_open_keyword()" (which is used to compute an error) because it relies on the stack of visited nodes. We cannot modify a stack of node references while we do the "mut" walk. Happily, an idiomatic solution is easy: we can tell the AST visitor to backtrack to the parent node and create the error there. Since "node_t::accept_base" is no longer a template we don't need the "node_visitation_t" trampoline anymore. The added copying at the FFI boundary makes things slower (memcpy dominates the profile) but it's not unusable, which is good news: $ hyperfine ./fish.{old,new}" -c 'source ../share/completions/git.fish'" Benchmark 1: ./fish.old -c 'source ../share/completions/git.fish' Time (mean ± σ): 195.5 ms ± 2.9 ms [User: 190.1 ms, System: 4.4 ms] Range (min … max): 193.2 ms … 205.1 ms 15 runs Benchmark 2: ./fish.new -c 'source ../share/completions/git.fish' Time (mean ± σ): 677.5 ms ± 62.0 ms [User: 665.4 ms, System: 10.0 ms] Range (min … max): 611.7 ms … 805.5 ms 10 runs Summary './fish.old -c 'source ../share/completions/git.fish'' ran 3.47 ± 0.32 times faster than './fish.new -c 'source ../share/completions/git.fish'' Leftovers: - Enum variants are still snakecase; I didn't get around to changing this yet. - "ast_type_to_string()" still returns a snakecase name. This could be changed since it's not user visible.
2023-04-02 14:42:59 +00:00
#include "fish_indent_common.h"
#include "ast.h"
#include "common.h"
#include "env.h"
#include "expand.h"
#include "flog.h"
#include "global_safety.h"
#include "maybe.h"
#include "operation_context.h"
#include "parse_constants.h"
#include "parse_util.h"
#include "tokenizer.h"
#include "wcstringutil.h"
#if INCLUDE_RUST_HEADERS
#include "fish_indent.rs.h"
#endif
using namespace ast;
// The number of spaces per indent isn't supposed to be configurable.
// See discussion at https://github.com/fish-shell/fish-shell/pull/6790
#define SPACES_PER_INDENT 4
/// \return whether a character at a given index is escaped.
/// A character is escaped if it has an odd number of backslashes.
static bool char_is_escaped(const wcstring &text, size_t idx) {
return count_preceding_backslashes(text, idx) % 2 == 1;
}
pretty_printer_t::pretty_printer_t(const wcstring &src, bool do_indent)
: source(src),
indents(do_indent ? parse_util_compute_indents(source) : std::vector<int>(src.size(), 0)),
ast(ast_parse(src, parse_flags())),
visitor(new_pretty_printer(*this)),
do_indent(do_indent),
gaps(compute_gaps()),
preferred_semi_locations(compute_preferred_semi_locations()) {
assert(indents.size() == source.size() && "indents and source should be same length");
}
pretty_printer_t::gap_flags_t pretty_printer_t::gap_text_flags_before_node(const node_t &node) {
gap_flags_t result = default_flags;
switch (node.typ()) {
// Allow escaped newlines before leaf nodes that can be part of a long command.
case type_t::argument:
case type_t::redirection:
case type_t::variable_assignment:
result |= allow_escaped_newlines;
break;
case type_t::token_base:
// Allow escaped newlines before && and ||, and also pipes.
switch (node.token_type()) {
case parse_token_type_t::andand:
case parse_token_type_t::oror:
case parse_token_type_t::pipe:
result |= allow_escaped_newlines;
break;
case parse_token_type_t::string: {
// Allow escaped newlines before commands that follow a variable assignment
// since both can be long (#7955).
auto p = node.parent();
if (p->typ() != type_t::decorated_statement) break;
p = p->parent();
assert(p->typ() == type_t::statement);
p = p->parent();
if (auto *job = p->try_as_job_pipeline()) {
if (!job->variables().empty()) result |= allow_escaped_newlines;
} else if (auto *job_cnt = p->try_as_job_continuation()) {
if (!job_cnt->variables().empty()) result |= allow_escaped_newlines;
} else if (auto *not_stmt = p->try_as_not_statement()) {
if (!not_stmt->variables().empty()) result |= allow_escaped_newlines;
}
break;
}
default:
break;
}
break;
default:
break;
}
return result;
}
bool pretty_printer_t::has_preceding_space() const {
long idx = static_cast<long>(output.size()) - 1;
// Skip escaped newlines.
// This is historical. Example:
//
// cmd1 \
// | cmd2
//
// we want the pipe to "see" the space after cmd1.
// TODO: this is too tricky, we should factor this better.
while (idx >= 0 && output.at(idx) == L'\n') {
size_t backslashes = count_preceding_backslashes(source, idx);
if (backslashes % 2 == 0) {
// Not escaped.
return false;
}
idx -= (1 + backslashes);
}
return idx >= 0 && output.at(idx) == L' ' && !char_is_escaped(output, idx);
}
wcstring pretty_printer_t::prettify() {
output = wcstring{};
visitor->visit(*ast->top());
// Trailing gap text.
emit_gap_text_before(source_range_t{(uint32_t)source.size(), 0}, default_flags);
// Replace all trailing newlines with just a single one.
while (!output.empty() && at_line_start()) {
output.pop_back();
}
emit_newline();
wcstring result = std::move(output);
return result;
}
std::vector<source_range_t> pretty_printer_t::compute_gaps() const {
auto range_compare = [](source_range_t r1, source_range_t r2) {
if (r1.start != r2.start) return r1.start < r2.start;
return r1.length < r2.length;
};
// Collect the token ranges into a list.
std::vector<source_range_t> tok_ranges;
for (auto ast_traversal = new_ast_traversal(*ast->top());;) {
auto node = ast_traversal->next();
if (!node->has_value()) break;
if (node->category() == category_t::leaf) {
auto r = node->source_range();
if (r.length > 0) tok_ranges.push_back(r);
}
}
// Place a zero length range at end to aid in our inverting.
tok_ranges.push_back(source_range_t{(uint32_t)source.size(), 0});
// Our tokens should be sorted.
assert(std::is_sorted(tok_ranges.begin(), tok_ranges.end(), range_compare));
// For each range, add a gap range between the previous range and this range.
std::vector<source_range_t> gaps;
uint32_t prev_end = 0;
for (source_range_t tok_range : tok_ranges) {
assert(tok_range.start >= prev_end && "Token range should not overlap or be out of order");
if (tok_range.start >= prev_end) {
gaps.push_back(source_range_t{prev_end, tok_range.start - prev_end});
}
prev_end = tok_range.start + tok_range.length;
}
return gaps;
}
void pretty_printer_t::visit_begin_header() {
if (!at_line_start()) {
emit_newline();
}
}
void pretty_printer_t::visit_maybe_newlines(const void *node_) {
const auto &node = *static_cast<const maybe_newlines_t *>(node_);
// Our newlines may have comments embedded in them, example:
// cmd |
// # something
// cmd2
// Treat it as gap text.
if (node.range().length > 0) {
auto flags = gap_text_flags_before_node(*node.ptr());
current_indent = indents.at(node.range().start);
bool added_newline = emit_gap_text_before(node.range(), flags);
source_range_t gap_range = node.range();
if (added_newline && gap_range.length > 0 && source.at(gap_range.start) == L'\n') {
gap_range.start++;
}
emit_gap_text(gap_range, flags);
}
}
void pretty_printer_t::visit_redirection(const void *node_) {
const auto &node = *static_cast<const redirection_t *>(node_);
// No space between a redirection operator and its target (#2899).
emit_text(node.oper().range(), default_flags);
emit_text(node.target().range(), skip_space);
}
void pretty_printer_t::visit_semi_nl(const void *node_) {
// These are semicolons or newlines which are part of the ast. That means it includes e.g.
// ones terminating a job or 'if' header, but not random semis in job lists. We respect
// preferred_semi_locations to decide whether or not these should stay as newlines or
// become semicolons.
const auto &node = *static_cast<const node_t *>(node_);
auto range = node.source_range();
// Check if we should prefer a semicolon.
bool prefer_semi =
range.length > 0 && std::binary_search(preferred_semi_locations.begin(),
preferred_semi_locations.end(), range.start);
emit_gap_text_before(range, gap_text_flags_before_node(*node.ptr()));
// Don't emit anything if the gap text put us on a newline (because it had a comment).
if (!at_line_start()) {
prefer_semi ? emit_semi() : emit_newline();
// If it was a semi but we emitted a newline, swallow a subsequent newline.
if (!prefer_semi && substr(range) == L";") {
gap_text_mask_newline = true;
}
}
}
void pretty_printer_t::emit_node_text(const void *node_) {
const auto &node = *static_cast<const node_t *>(node_);
source_range_t range = node.source_range();
// Weird special-case: a token may end in an escaped newline. Notably, the newline is
// not part of the following gap text, handle indentation here (#8197).
bool ends_with_escaped_nl = range.length >= 2 && source.at(range.end() - 2) == L'\\' &&
source.at(range.end() - 1) == L'\n';
if (ends_with_escaped_nl) {
range = {range.start, range.length - 2};
}
emit_text(range, gap_text_flags_before_node(node));
if (ends_with_escaped_nl) {
// By convention, escaped newlines are preceded with a space.
output.append(L" \\\n");
// TODO Maybe check "allow_escaped_newlines" and use the precomputed indents.
// The cases where this matters are probably very rare.
current_indent++;
emit_space_or_indent();
current_indent--;
}
}
void pretty_printer_t::emit_text(source_range_t r, gap_flags_t flags) {
emit_gap_text_before(r, flags);
current_indent = indents.at(r.start);
if (r.length > 0) {
emit_space_or_indent(flags);
output.append(clean_text(substr(r)));
}
}
wcstring pretty_printer_t::clean_text(const wcstring &input) {
// Unescape the string - this leaves special markers around if there are any
// expansions or anything. We specifically tell it to not compute backslash-escapes
// like \U or \x, because we want to leave them intact.
wcstring unescaped = input;
unescape_string_in_place(&unescaped, UNESCAPE_SPECIAL | UNESCAPE_NO_BACKSLASHES);
// Remove INTERNAL_SEPARATOR because that's a quote.
auto quote = [](wchar_t ch) { return ch == INTERNAL_SEPARATOR; };
unescaped.erase(std::remove_if(unescaped.begin(), unescaped.end(), quote), unescaped.end());
// If no non-"good" char is left, use the unescaped version.
// This can be extended to other characters, but giving the precise list is tough,
// can change over time (see "^", "%" and "?", in some cases "{}") and it just makes
// people feel more at ease.
auto goodchars = [](wchar_t ch) {
return fish_iswalnum(ch) || ch == L'_' || ch == L'-' || ch == L'/';
};
if (std::find_if_not(unescaped.begin(), unescaped.end(), goodchars) == unescaped.end() &&
!unescaped.empty()) {
return unescaped;
} else {
return input;
}
}
bool pretty_printer_t::emit_gap_text_before(source_range_t r, gap_flags_t flags) {
assert(r.start <= source.size() && "source out of bounds");
bool added_newline = false;
// Find the gap text which ends at start.
source_range_t range = gap_text_to(r.start);
if (range.length > 0) {
// Set the indent from the beginning of this gap text.
// For example:
// begin
// cmd
// # comment
// end
// Here the comment is the gap text before the end, but we want the indent from the
// command.
if (range.start < indents.size()) current_indent = indents.at(range.start);
// If this range contained an error, append the gap text without modification.
// For example in: echo foo "
// We don't want to mess with the quote.
if (range_contained_error(range)) {
output.append(substr(range));
} else {
added_newline = emit_gap_text(range, flags);
}
}
// Always clear gap_text_mask_newline after emitting even empty gap text.
gap_text_mask_newline = false;
return added_newline;
}
bool pretty_printer_t::range_contained_error(source_range_t r) const {
const auto &errs = ast->extras()->errors();
auto range_is_before = [](source_range_t x, source_range_t y) {
return x.start + x.length <= y.start;
};
assert(std::is_sorted(errs.begin(), errs.end(), range_is_before) &&
"Error ranges should be sorted");
return std::binary_search(errs.begin(), errs.end(), r, range_is_before);
}
source_range_t pretty_printer_t::gap_text_to(uint32_t end) const {
auto where =
std::lower_bound(gaps.begin(), gaps.end(), end,
[](source_range_t r, uint32_t end) { return r.start + r.length < end; });
if (where == gaps.end() || where->start + where->length != end) {
// Not found.
return source_range_t{0, 0};
} else {
return *where;
}
}
bool pretty_printer_t::emit_gap_text(source_range_t range, gap_flags_t flags) {
wcstring gap_text = substr(range);
// Common case: if we are only spaces, do nothing.
if (gap_text.find_first_not_of(L' ') == wcstring::npos) return false;
// Look to see if there is an escaped newline.
// Emit it if either we allow it, or it comes before the first comment.
// Note we do not have to be concerned with escaped backslashes or escaped #s. This is gap
// text - we already know it has no semantic significance.
size_t escaped_nl = gap_text.find(L"\\\n");
if (escaped_nl != wcstring::npos) {
size_t comment_idx = gap_text.find(L'#');
if ((flags & allow_escaped_newlines) ||
(comment_idx != wcstring::npos && escaped_nl < comment_idx)) {
// Emit a space before the escaped newline.
if (!at_line_start() && !has_preceding_space()) {
output.append(L" ");
}
output.append(L"\\\n");
// Indent the continuation line and any leading comments (#7252).
// Use the indentation level of the next newline.
current_indent = indents.at(range.start + escaped_nl + 1);
emit_space_or_indent();
}
}
// It seems somewhat ambiguous whether we always get a newline after a comment. Ensure we
// always emit one.
bool needs_nl = false;
auto tokenizer = new_tokenizer(gap_text.c_str(), TOK_SHOW_COMMENTS | TOK_SHOW_BLANK_LINES);
while (auto tok = tokenizer->next()) {
wcstring tok_text = *tokenizer->text_of(*tok);
if (needs_nl) {
emit_newline();
needs_nl = false;
if (tok_text == L"\n") continue;
} else if (gap_text_mask_newline) {
// We only respect mask_newline the first time through the loop.
gap_text_mask_newline = false;
if (tok_text == L"\n") continue;
}
if (tok->type_ == token_type_t::comment) {
emit_space_or_indent();
output.append(tok_text);
needs_nl = true;
} else if (tok->type_ == token_type_t::end) {
// This may be either a newline or semicolon.
// Semicolons found here are not part of the ast and can simply be removed.
// Newlines are preserved unless mask_newline is set.
if (tok_text == L"\n") {
emit_newline();
}
} else {
fprintf(stderr,
"Gap text should only have comments and newlines - instead found token "
"type %d with text: %ls\n",
(int)tok->type_, tok_text.c_str());
DIE("Gap text should only have comments and newlines");
}
}
if (needs_nl) emit_newline();
return needs_nl;
}
void pretty_printer_t::emit_space_or_indent(gap_flags_t flags) {
if (at_line_start()) {
output.append(SPACES_PER_INDENT * current_indent, L' ');
} else if (!(flags & skip_space) && !has_preceding_space()) {
output.append(1, L' ');
}
}
std::vector<uint32_t> pretty_printer_t::compute_preferred_semi_locations() const {
std::vector<uint32_t> result;
auto mark_semi_from_input = [&](const semi_nl_t &n) {
if (n.ptr()->has_source() && substr(n.range()) == L";") {
result.push_back(n.range().start);
}
};
// andor_job_lists get semis if the input uses semis.
for (auto ast_traversal = new_ast_traversal(*ast->top());;) {
auto node = ast_traversal->next();
if (!node->has_value()) break;
// See if we have a condition and an andor_job_list.
const semi_nl_t *condition = nullptr;
const andor_job_list_t *andors = nullptr;
if (const auto *ifc = node->try_as_if_clause()) {
if (ifc->condition().has_semi_nl()) {
condition = &ifc->condition().semi_nl();
}
andors = &ifc->andor_tail();
} else if (const auto *wc = node->try_as_while_header()) {
if (wc->condition().has_semi_nl()) {
condition = &wc->condition().semi_nl();
}
andors = &wc->andor_tail();
}
// If there is no and-or tail then we always use a newline.
if (andors && andors->count() > 0) {
if (condition) mark_semi_from_input(*condition);
// Mark all but last of the andor list.
for (uint32_t i = 0; i + 1 < andors->count(); i++) {
mark_semi_from_input(andors->at(i)->job().semi_nl());
}
}
}
// `x ; and y` gets semis if it has them already, and they are on the same line.
for (auto ast_traversal = new_ast_traversal(*ast->top());;) {
auto node = ast_traversal->next();
if (!node->has_value()) break;
if (const auto *job_list = node->try_as_job_list()) {
const semi_nl_t *prev_job_semi_nl = nullptr;
for (size_t i = 0; i < job_list->count(); i++) {
const job_conjunction_t &job = *job_list->at(i);
// Set up prev_job_semi_nl for the next iteration to make control flow easier.
const semi_nl_t *prev = prev_job_semi_nl;
prev_job_semi_nl = job.has_semi_nl() ? &job.semi_nl() : nullptr;
// Is this an 'and' or 'or' job?
if (!job.has_decorator()) continue;
// Now see if we want to mark 'prev' as allowing a semi.
// Did we have a previous semi_nl which was a newline?
if (!prev || substr(prev->range()) != L";") continue;
// Is there a newline between them?
assert(prev->range().start <= job.decorator().range().start &&
"Ranges out of order");
auto start = source.begin() + prev->range().start;
auto end = source.begin() + job.decorator().range().end();
if (std::find(start, end, L'\n') == end) {
// We're going to allow the previous semi_nl to be a semi.
result.push_back(prev->range().start);
}
}
}
}
std::sort(result.begin(), result.end());
return result;
}