// SPDX-License-Identifier: GPL-3.0-or-later

package engine

import (
	"context"
	"errors"
	"fmt"
	"io/fs"
	"path/filepath"
	"sort"
	"strings"
	"sync"
	"time"

	"git.labunix.xyz/krino/internal/cond"
	"git.labunix.xyz/krino/internal/dup"
	"git.labunix.xyz/krino/internal/extract"
	"git.labunix.xyz/krino/internal/kwcache"
	"git.labunix.xyz/krino/internal/norm"
	"git.labunix.xyz/krino/internal/plan"
	"git.labunix.xyz/krino/internal/scan"
	"git.labunix.xyz/krino/internal/xdg"
)

// matchRun holds the state shared by every file evaluated during one Match
// or Explain call: the directory being matched, the full set of scanned
// files (for duplicate detection) and the lazily built duplicate indexes,
// one per distinct set of resolved directories a (duplicate ...) test
// names. mu guards dupOnce, dupIdx and warn, the only fields any goroutine
// but the one that created the matchRun ever touches.
type matchRun struct {
	e     *Engine
	d     *Dir
	ctx   context.Context
	now   time.Time
	files []scan.File
	cache *kwcache.Cache // nil: no keyword cache

	// bigText limits how many large files have their text in memory at
	// once. Extracting is otherwise GOMAXPROCS-wide, and a large file's
	// text costs several times its own size: it is read whole, a file that
	// is not valid UTF-8 doubles as it decodes, and each distinct set of
	// case/fold options makes another normalised copy. Sixteen 49 MB text
	// files reached 2.7 GB. Small files are unaffected, which is nearly
	// every file.
	bigText chan struct{}

	mu      sync.Mutex
	dupOnce map[string]*sync.Once
	dupIdx  map[string]*dup.Index
	warn    []string
}

// bigTextSize is the file size above which extraction is rationed, and
// bigTextAtOnce is how many such files may be in flight together.
const (
	bigTextSize   = 4 << 20 // 4 MiB
	bigTextAtOnce = 2
)

// newMatchRun builds a matchRun over files, the set a (duplicate ...) test
// with no directories of its own checks against.
func newMatchRun(e *Engine, d *Dir, ctx context.Context, now time.Time, files []scan.File) *matchRun {
	return &matchRun{
		e:       e,
		d:       d,
		ctx:     ctx,
		now:     now,
		files:   files,
		dupOnce: make(map[string]*sync.Once),
		dupIdx:  make(map[string]*dup.Index),
		bigText: make(chan struct{}, bigTextAtOnce),
	}
}

// warnings returns the directory-level warnings collected so far (from
// building duplicate indexes), in the order they were recorded.
func (run *matchRun) warnings() []string {
	run.mu.Lock()
	defer run.mu.Unlock()
	return append([]string(nil), run.warn...)
}

// drainDupErrors appends every duplicate index's candidate-hashing errors
// (a candidate other than the file being looked up that could not be
// hashed) to run.warn, once matching is done and every index has seen every
// Lookup it is going to see. Candidate paths are abbreviated with
// xdg.Abbrev, as every other user-visible path is.
func (run *matchRun) drainDupErrors() {
	run.mu.Lock()
	defer run.mu.Unlock()
	for _, idx := range run.dupIdx {
		for _, ce := range idx.Errors() {
			run.warn = append(run.warn, dupWarning(ce))
		}
	}
}

// dupWarning is the one form of a duplicate warning: "duplicate: PATH:
// cause", the path abbreviated like every other one shown, and an OS
// error's cause without the raw path it would repeat.
func dupWarning(err error) string {
	return "duplicate: " + dupCause(err).Error()
}

// dupCause is a duplicate check's error as "PATH: cause", the path
// shortened with xdg.Abbrev and not repeated inside the cause - a candidate
// that could not be hashed, or an OS error on the file itself. It still
// unwraps to err.
func dupCause(err error) error {
	path, cause := "", err
	var ce dup.CandidateError
	var pe *fs.PathError
	switch {
	case errors.As(err, &ce):
		path, cause = ce.Path, ce.Err
		if errors.As(ce.Err, &pe) {
			cause = fmt.Errorf("%s: %w", pe.Op, pe.Err)
		}
	case errors.As(err, &pe):
		path, cause = pe.Path, fmt.Errorf("%s: %w", pe.Op, pe.Err)
	default:
		return err
	}
	return shortenedErr{msg: xdg.Abbrev(path) + ": " + cause.Error(), err: err}
}

type shortenedErr struct {
	msg string
	err error
}

func (e shortenedErr) Error() string { return e.msg }
func (e shortenedErr) Unwrap() error { return e.err }

// dupIndex returns the shared *dup.Index for the resolved, sorted extra
// directories named by key, building it exactly once across every
// concurrent caller that asks for the same key.
func (run *matchRun) dupIndex(key string, dirs []string) *dup.Index {
	run.mu.Lock()
	once, ok := run.dupOnce[key]
	if !ok {
		once = &sync.Once{}
		run.dupOnce[key] = once
	}
	run.mu.Unlock()

	once.Do(func() {
		idx, errs := dup.NewIndex(run.files, dirs)
		run.mu.Lock()
		run.dupIdx[key] = idx
		for _, err := range errs {
			run.warn = append(run.warn, dupWarning(err))
		}
		run.mu.Unlock()
	})

	run.mu.Lock()
	idx := run.dupIdx[key]
	run.mu.Unlock()
	return idx
}

// facts is one file's cond.Facts. It is used by exactly one goroutine, so
// its own memoised state (the keyword answers, and whether an earlier rule
// matched) needs no locking of its own; only the matchRun it points at is
// shared.
type facts struct {
	run  *matchRun
	file scan.File

	matched bool
	// undecided: an earlier rule's condition was unknown (content krino
	// could not read), so (matched) is unknown while none has matched.
	undecided bool

	// dupOriginal is the file a (duplicate) test last matched against,
	// absolute, for a front end that offers to act on the other copy.
	dupOriginal string

	// folded memoises the folded form of this file's name and path, which
	// every name test of every rule would otherwise recompute.
	folded map[string]norm.Folded

	contentDone bool            // extraction was attempted
	contentErr  error           // why it failed
	partialErr  error           // extract.ErrPartial: a keyword not found may be in the unread part
	answers     map[string]bool // by cond.KeywordKey, once extracted
}

var _ cond.Facts = (*facts)(nil)

// newFacts builds the Facts for one scanned file.
func newFacts(run *matchRun, file scan.File) *facts {
	return &facts{run: run, file: file}
}

// Folded is the file's name or path folded for comparison, worked out once
// however many name tests ask for it. facts belongs to one file and one
// goroutine, so no lock is needed.
func (f *facts) Folded(subj string) norm.Folded {
	if v, ok := f.folded[subj]; ok {
		return v
	}
	v := norm.FoldMapped(subj)
	if f.folded == nil {
		f.folded = make(map[string]norm.Folded, 2)
	}
	f.folded[subj] = v
	return v
}

func (f *facts) Name() string          { return f.file.Name }
func (f *facts) Rel() string           { return f.file.Rel }
func (f *facts) Size() int64           { return f.file.Size }
func (f *facts) ModTime() time.Time    { return f.file.ModTime }
func (f *facts) Now() time.Time        { return f.run.now }
func (f *facts) Matched() (bool, bool) { return f.matched, f.undecided }

// ContentContains answers a content test (spec §6.1). A file above
// max-read is never read, cached or not. Before the file has been
// extracted this run, the keyword cache answers when it knows every one of
// keywords for this file as it is now; otherwise the text is extracted
// once, every keyword of the directory (and of this test) is answered from
// it and stored in the cache, and the text itself is dropped. A failed
// extraction is not cached: the next run tries again.
func (f *facts) ContentContains(opt cond.Options, keywords []string) (int, error) {
	if max := f.run.d.Settings.MaxRead; max > 0 && f.file.Size > max {
		return -1, extract.ErrTooLarge
	}
	keys := make([]string, len(keywords))
	for i, kw := range keywords {
		keys[i] = cond.KeywordKey(opt, kw)
	}
	if !f.contentDone {
		if id, ok := f.cacheID(); ok {
			if hits, ok := f.run.cache.Lookup(id, keys); ok {
				for i, hit := range hits {
					if hit {
						return i, nil
					}
				}
				return -1, nil
			}
		}
		f.extract(opt, keywords)
	}
	if f.contentErr != nil {
		return -1, f.contentErr
	}
	for i, k := range keys {
		if f.answers[k] {
			return i, nil
		}
	}
	return -1, f.partialErr
}

// extract reads the file's text and answers every keyword of the directory,
// plus the asking test's (opt, keywords), from it.
func (f *facts) extract(opt cond.Options, keywords []string) {
	f.contentDone = true
	if f.file.Size >= bigTextSize && f.run.bigText != nil {
		// Wait for a slot rather than hold several large files' text at
		// once. Cancellation is still noticed: Text takes the same ctx.
		select {
		case f.run.bigText <- struct{}{}:
			defer func() { <-f.run.bigText }()
		case <-f.run.ctx.Done():
		}
	}
	text, err := f.run.e.Extract.Text(f.run.ctx, f.file.Path, f.file.Size, f.run.d.Settings.MaxRead)
	switch {
	case errors.Is(err, extract.ErrPartial):
		// Partly read: the keywords found in it are answered; one not found
		// is unknown (ContentContains), and nothing is cached.
		f.partialErr = err
	case errors.Is(err, extract.ErrUnsupported):
		// A format with no text cannot contain a keyword: every answer is
		// no, with no warning, and the answers are cached like any other.
		// Only a real read failure is "unreadable".
		text = ""
	case err != nil:
		f.contentErr = err
		return
	}
	all := append([]cond.Keyword(nil), f.run.d.ContentKeywords...)
	for _, kw := range keywords {
		all = append(all, cond.Keyword{Opt: opt, Norm: kw})
	}
	normed := map[cond.Options]string{}
	f.answers = make(map[string]bool, len(all))
	for _, k := range all {
		t, ok := normed[k.Opt]
		if !ok {
			t = norm.Text(text, k.Opt.IgnoreCase, k.Opt.Fold)
			normed[k.Opt] = t
		}
		f.answers[k.Key()] = strings.Contains(t, k.Norm)
	}
	if id, ok := f.cacheID(); ok && f.partialErr == nil {
		f.run.cache.Store(id, f.answers)
	}
}

// cacheID is the file's keyword cache identity; ok is false when there is
// no cache, or the platform gave the file no inode.
func (f *facts) cacheID() (kwcache.ID, bool) {
	if f.run.cache == nil || f.file.Ino == 0 {
		return kwcache.ID{}, false
	}
	return fileCacheID(f.file), true
}

// fileCacheID is file's kwcache.ID.
func fileCacheID(file scan.File) kwcache.ID {
	return kwcache.ID{Dev: file.Dev, Ino: file.Ino, Size: file.Size, MTime: file.ModTime.UnixNano(), Ext: strings.ToLower(filepath.Ext(file.Name))}
}

// Duplicate resolves dirs against the directory's root, builds (or reuses)
// the shared duplicate index for that resolved, sorted set, and looks the
// file up in it.
func (f *facts) Duplicate(dirs []string) (string, bool, error) {
	root := f.run.d.Root
	resolved := make([]string, len(dirs))
	for i, raw := range dirs {
		resolved[i] = plan.ResolveDir(raw, root)
	}
	sorted := append([]string(nil), resolved...)
	sort.Strings(sorted)
	key := strings.Join(sorted, "\x00")

	idx := f.run.dupIndex(key, sorted)
	orig, isDup, err := idx.Lookup(f.file.Path)
	if err != nil {
		return "", false, dupCause(err)
	}
	if !isDup {
		return "", false, nil
	}
	// The absolute path is kept for a front end that has to act on the
	// other copy - the window offers to keep this one instead - while the
	// reason text stays as it reads best.
	f.dupOriginal = orig
	return displayOriginal(orig, root), true, nil
}

// DuplicateOriginal is the file the last (duplicate) test matched against,
// absolute; "" when none did.
func (f *facts) DuplicateOriginal() string { return f.dupOriginal }

// displayOriginal reports orig relative to root when it lies inside root,
// else home-abbreviated (xdg.Abbrev), as every other user-visible path is.
func displayOriginal(orig, root string) string {
	rel, err := filepath.Rel(root, orig)
	if err != nil || rel == ".." || strings.HasPrefix(rel, ".."+string(filepath.Separator)) {
		return xdg.Abbrev(orig)
	}
	return filepath.ToSlash(rel)
}