QuoteForge/internal/article/generator.go

package article

import (
	"fmt"
	"regexp"
	"sort"
	"strings"

	"git.mchus.pro/mchus/quoteforge/internal/localdb"
	"git.mchus.pro/mchus/quoteforge/internal/models"
)

type BuildOptions struct {
	ServerModel string
}

type BuildResult struct {
	Article  string
	Segments []ResultSegment
	Warnings []string
}

// ResultSegment is one dash-separated piece of the article, tagged so the UI can
// highlight pieces the generator could not fully parse. Recognized is false when
// the segment contains a category placeholder (e.g. "GPU" instead of a real model
// token) because a lot_name did not match the expected naming shape.
type ResultSegment struct {
	Group      string `json:"group"`
	Text       string `json:"text"`
	Recognized bool   `json:"recognized"`
}

var (
	reMemGiB    = regexp.MustCompile(`(?i)(\d+)\s*(GB|G)`)
	reMemTiB    = regexp.MustCompile(`(?i)(\d+)\s*(TB|T)`)
	reCapacityT = regexp.MustCompile(`(?i)(\d+(?:[.,]\d+)?)T`)
	reCapacityG = regexp.MustCompile(`(?i)(\d+(?:[.,]\d+)?)G`)
	rePortSpeed = regexp.MustCompile(`(?i)(\d+)p(\d+)(GbE|G)`)
	rePortFC    = regexp.MustCompile(`(?i)(\d+)pFC(\d+)`)
	reWatts     = regexp.MustCompile(`(?i)(\d{3,5})\s*W`)
)

type namedSeg struct {
	group string // "MODEL","CPU","MEM","GPU","DISK","NET","PSU","SUPPORT"
	value string
	// degraded marks a segment whose value contains a category placeholder instead
	// of a parsed spec token, because some lot_name did not match the expected shape.
	degraded bool
}

// segmentResult is what each build*Segment helper returns: the rendered value, a
// degraded flag, and human-readable warnings that name the offending lot_names.
type segmentResult struct {
	value    string
	degraded bool
	warnings []string
}

func Build(local *localdb.LocalDB, items []models.ConfigItem, opts BuildOptions) (BuildResult, error) {
	segs := make([]namedSeg, 0, 8)
	warnings := make([]string, 0)

	model := NormalizeServerModel(opts.ServerModel)
	if model == "" {
		return BuildResult{}, fmt.Errorf("server_model required")
	}
	segs = append(segs, namedSeg{group: "MODEL", value: model})

	lotNames := make([]string, 0, len(items))
	for _, it := range items {
		lotNames = append(lotNames, it.LotName)
	}

	cats, err := ResolveLotCategories(local, lotNames)
	if err != nil {
		return BuildResult{}, err
	}

	for _, sb := range []struct {
		group string
		build func([]models.ConfigItem, map[string]string) segmentResult
	}{
		{"CPU", buildCPUSegment},
		{"MEM", buildMemSegment},
		{"GPU", buildGPUSegment},
		{"DISK", buildDiskSegment},
		{"NET", buildNetSegment},
		{"PSU", buildPSUSegment},
	} {
		res := sb.build(items, cats)
		warnings = append(warnings, res.warnings...)
		if res.value != "" {
			segs = append(segs, namedSeg{group: sb.group, value: res.value, degraded: res.degraded})
		}
	}

	if supportSeg := buildSupportSegment(items); supportSeg != "" {
		segs = append(segs, namedSeg{group: "SUPPORT", value: supportSeg})
	}

	article := strings.Join(namedSegsValues(segs), "-")
	if len([]rune(article)) > 80 {
		segs = compressArticle(segs)
		article = strings.Join(namedSegsValues(segs), "-")
		warnings = append(warnings, "compressed")
	}
	if len([]rune(article)) > 80 {
		return BuildResult{}, fmt.Errorf("article_overflow")
	}

	result := BuildResult{Article: article, Warnings: warnings}
	for _, s := range segs {
		result.Segments = append(result.Segments, ResultSegment{
			Group:      s.group,
			Text:       s.value,
			Recognized: !s.degraded,
		})
	}
	return result, nil
}

func namedSegsValues(segs []namedSeg) []string {
	out := make([]string, len(segs))
	for i, s := range segs {
		out[i] = s.value
	}
	return out
}

func findSegGroup(segs []namedSeg, group string) int {
	for i, s := range segs {
		if s.group == group {
			return i
		}
	}
	return -1
}

// buildSupportSegment finds a support LOT in items (added to the spec via the
// Base tab's support picker, e.g. "SVC_3yB_HGX-H200") and returns its
// duration+level token ("3yB") for the article. Support is a regular BOM
// LOT, not pricelist-backed, so it's detected by lot_name prefix like the
// other lot_name-pattern parsers in this file, rather than by lot_category.
func buildSupportSegment(items []models.ConfigItem) string {
	for _, it := range items {
		if !strings.HasPrefix(strings.ToUpper(it.LotName), "SVC_") {
			continue
		}
		parts := strings.SplitN(it.LotName, "_", 3)
		if len(parts) >= 2 && parts[1] != "" {
			return parts[1]
		}
	}
	return ""
}

// placeholderToken returns the token used in the article when a lot's spec can't be
// parsed: the lot_category itself (never a bare "UNK"), uppercased.
func placeholderToken(cat string) string {
	t := strings.ToUpper(strings.TrimSpace(cat))
	if t == "" {
		return "X"
	}
	return t
}

func buildCPUSegment(items []models.ConfigItem, cats map[string]string) segmentResult {
	models := map[string]int{}
	res := segmentResult{}
	for _, it := range items {
		group, ok := GroupForLotCategory(cats[it.LotName])
		if !ok || group != GroupCPU {
			continue
		}
		model, parsed := parseCPUModel(it.LotName)
		if !parsed {
			model = placeholderToken(cats[it.LotName])
			res.degraded = true
			res.warnings = append(res.warnings, fmt.Sprintf("CPU: не распознана модель LOT %q — в артикул записана категория %q", it.LotName, model))
		}
		models[model] += it.Quantity
	}
	if len(models) == 0 {
		return res
	}
	res.value = joinQtyTokens(models)
	return res
}

func buildMemSegment(items []models.ConfigItem, cats map[string]string) segmentResult {
	totalGiB := 0
	res := segmentResult{}
	unparsed := 0
	for _, it := range items {
		group, ok := GroupForLotCategory(cats[it.LotName])
		if !ok || group != GroupMEM {
			continue
		}
		per := parseMemGiB(it.LotName)
		if per <= 0 {
			unparsed++
			res.degraded = true
			res.warnings = append(res.warnings, fmt.Sprintf("MEM: не распознан объём LOT %q", it.LotName))
			continue
		}
		totalGiB += per * it.Quantity
	}
	parts := make([]string, 0, 2)
	if totalGiB > 0 {
		if totalGiB%1024 == 0 {
			parts = append(parts, fmt.Sprintf("%dT", totalGiB/1024))
		} else {
			parts = append(parts, fmt.Sprintf("%dG", totalGiB))
		}
	}
	if unparsed > 0 {
		parts = append(parts, placeholderToken("MEM"))
	}
	res.value = strings.Join(parts, "+")
	return res
}

func buildGPUSegment(items []models.ConfigItem, cats map[string]string) segmentResult {
	models := map[string]int{}
	res := segmentResult{}
	for _, it := range items {
		group, ok := GroupForLotCategory(cats[it.LotName])
		if !ok || group != GroupGPU {
			continue
		}
		if strings.HasPrefix(strings.ToUpper(it.LotName), "MB_") {
			continue
		}
		model, parsed := parseGPUModel(it.LotName)
		if !parsed {
			model = placeholderToken(cats[it.LotName])
			res.degraded = true
			res.warnings = append(res.warnings, fmt.Sprintf("GPU: не распознана модель LOT %q — в артикул записана категория %q", it.LotName, model))
		}
		models[model] += it.Quantity
	}
	if len(models) == 0 {
		return res
	}
	res.value = joinQtyTokens(models)
	return res
}

func buildDiskSegment(items []models.ConfigItem, cats map[string]string) segmentResult {
	type key struct {
		t string
		c string
	}
	groupQty := map[key]int{}
	res := segmentResult{}
	for _, it := range items {
		group, ok := GroupForLotCategory(cats[it.LotName])
		if !ok || group != GroupDISK {
			continue
		}
		capToken := parseCapacity(it.LotName)
		if capToken == "" {
			res.degraded = true
			res.warnings = append(res.warnings, fmt.Sprintf("DISK: не распознан объём LOT %q", it.LotName))
		}
		typeCode := diskTypeCode(cats[it.LotName], it.LotName)
		groupQty[key{t: typeCode, c: capToken}] += it.Quantity
	}
	if len(groupQty) == 0 {
		return res
	}
	parts := make([]string, 0, len(groupQty))
	for k, qty := range groupQty {
		if k.c == "" {
			parts = append(parts, fmt.Sprintf("%dx%s", qty, k.t))
		} else {
			parts = append(parts, fmt.Sprintf("%dx%s%s", qty, k.c, k.t))
		}
	}
	sort.Strings(parts)
	res.value = strings.Join(parts, "+")
	return res
}

func buildNetSegment(items []models.ConfigItem, cats map[string]string) segmentResult {
	return buildProfileSegment(items, cats, GroupNET, parsePortSpeed, "NET")
}

func buildPSUSegment(items []models.ConfigItem, cats map[string]string) segmentResult {
	return buildProfileSegment(items, cats, GroupPSU, parseWatts, "PSU")
}

// buildProfileSegment groups items of the given category group by a profile token
// parsed from their lot name (e.g. port speed, wattage rating). When a lot's profile
// can't be determined it falls back to the category placeholder and flags the segment
// as degraded, naming the lot in a warning.
func buildProfileSegment(items []models.ConfigItem, cats map[string]string, group Group, parseProfile func(string) string, groupLabel string) segmentResult {
	groupQty := map[string]int{}
	res := segmentResult{}
	for _, it := range items {
		g, ok := GroupForLotCategory(cats[it.LotName])
		if !ok || g != group {
			continue
		}
		profile := parseProfile(it.LotName)
		if profile == "" {
			profile = placeholderToken(groupLabel)
			res.degraded = true
			res.warnings = append(res.warnings, fmt.Sprintf("%s: не распознан профиль LOT %q — в артикул записана категория %q", groupLabel, it.LotName, profile))
		}
		groupQty[profile] += it.Quantity
	}
	if len(groupQty) == 0 {
		return res
	}
	res.value = joinQtyTokens(groupQty)
	return res
}

// joinQtyTokens renders {token: qty} as a sorted "NxTOKEN+MxTOKEN" string.
func joinQtyTokens(qty map[string]int) string {
	parts := make([]string, 0, len(qty))
	for token, n := range qty {
		parts = append(parts, fmt.Sprintf("%dx%s", n, token))
	}
	sort.Strings(parts)
	return strings.Join(parts, "+")
}

func normalizeModelToken(lotName string) string {
	if idx := strings.Index(lotName, "_"); idx >= 0 && idx+1 < len(lotName) {
		lotName = lotName[idx+1:]
	}
	parts := strings.Split(lotName, "_")
	token := parts[len(parts)-1]
	return strings.ToUpper(strings.TrimSpace(token))
}

// parseCPUModel extracts the model token from a CPU lot_name (shape
// CPU_{VENDOR}_{MODEL}, e.g. "CPU_INTEL_8592+"). The bool is false when the name
// has no parseable {VENDOR}_{MODEL} tail — a structural failure the caller surfaces.
// It cannot validate that the token is a real CPU model (no vendor catalog is kept
// in the repo by design), only that the name matched the expected shape.
func parseCPUModel(lotName string) (string, bool) {
	parts := strings.Split(lotName, "_")
	if len(parts) >= 2 {
		last := strings.ToUpper(strings.TrimSpace(parts[len(parts)-1]))
		if last != "" {
			return last, true
		}
	}
	return normalizeModelToken(lotName), false
}

// parseGPUModel extracts a "MODEL[_MEM]" token from a GPU lot_name. The bool is
// false when no model token could be located and the result is only the last-ditch
// last-underscore-segment fallback — a structural failure the caller surfaces.
// Like parseCPUModel it does not validate the token against a catalog.
func parseGPUModel(lotName string) (string, bool) {
	upper := strings.ToUpper(lotName)
	if idx := strings.Index(upper, "GPU_"); idx >= 0 {
		upper = upper[idx+4:]
	}
	parts := strings.Split(upper, "_")
	model := ""
	numSuffix := ""
	mem := ""
	for i, p := range parts {
		if p == "" {
			continue
		}
		switch p {
		case "NV", "NVIDIA", "INTEL", "AMD", "RADEON", "PCIE", "PCI", "SXM", "SXMX", "SFF", "LOVELACE":
			continue
		case "ADA", "AMPERE", "HOPPER", "BLACKWELL":
			if model != "" {
				archAbbr := map[string]string{
					"ADA": "ADA", "AMPERE": "AMP", "HOPPER": "HOP", "BLACKWELL": "BWL",
				}
				numSuffix += archAbbr[p]
			}
			continue
		default:
			if strings.Contains(p, "GB") {
				mem = p
				continue
			}
			if model == "" && i > 0 {
				model = p
			} else if model != "" && numSuffix == "" && isModelNumber(p) {
				numSuffix = p
			}
		}
	}
	full := model
	if numSuffix != "" {
		full = model + numSuffix
	}
	if full != "" && mem != "" {
		return full + "_" + mem, true
	}
	if full != "" {
		return full, true
	}
	return normalizeModelToken(lotName), false
}

// isModelNumber reports whether s looks like a GPU/accelerator model number token:
// it starts with a digit and contains only digits and uppercase letters. This keeps
// vendor-suffixed names like "6000D" (RTX PRO 6000D) intact instead of dropping the
// letter and collapsing two distinct models to the same token.
func isModelNumber(s string) bool {
	if s == "" || s[0] < '0' || s[0] > '9' {
		return false
	}
	for _, r := range s {
		if (r < '0' || r > '9') && (r < 'A' || r > 'Z') {
			return false
		}
	}
	return true
}

func parseMemGiB(lotName string) int {
	if m := reMemTiB.FindStringSubmatch(lotName); len(m) == 3 {
		return atoi(m[1]) * 1024
	}
	if m := reMemGiB.FindStringSubmatch(lotName); len(m) == 3 {
		return atoi(m[1])
	}
	return 0
}

func parseCapacity(lotName string) string {
	if m := reCapacityT.FindStringSubmatch(lotName); len(m) == 2 {
		return normalizeTToken(strings.ReplaceAll(m[1], ",", ".")) + "T"
	}
	if m := reCapacityG.FindStringSubmatch(lotName); len(m) == 2 {
		return normalizeNumberToken(strings.ReplaceAll(m[1], ",", ".")) + "G"
	}
	return ""
}

func diskTypeCode(cat string, lotName string) string {
	c := strings.ToUpper(strings.TrimSpace(cat))
	if c == "M2" {
		return "M2"
	}
	upper := strings.ToUpper(lotName)
	if strings.Contains(upper, "NVME") {
		return "NV"
	}
	if strings.Contains(upper, "SAS") {
		return "SAS"
	}
	if strings.Contains(upper, "SATA") {
		return "SAT"
	}
	return c
}

func parsePortSpeed(lotName string) string {
	if m := rePortSpeed.FindStringSubmatch(lotName); len(m) == 4 {
		return fmt.Sprintf("%sp%sG", m[1], m[2])
	}
	if m := rePortFC.FindStringSubmatch(lotName); len(m) == 3 {
		return fmt.Sprintf("%spFC%s", m[1], m[2])
	}
	return ""
}

func parseWatts(lotName string) string {
	if m := reWatts.FindStringSubmatch(lotName); len(m) == 2 {
		w := atoi(m[1])
		if w >= 1000 {
			kw := fmt.Sprintf("%.1f", float64(w)/1000.0)
			kw = strings.TrimSuffix(kw, ".0")
			return fmt.Sprintf("%skW", kw)
		}
		return fmt.Sprintf("%dW", w)
	}
	return ""
}

func normalizeNumberToken(raw string) string {
	raw = strings.TrimSpace(raw)
	raw = strings.TrimLeft(raw, "0")
	if raw == "" || raw[0] == '.' {
		raw = "0" + raw
	}
	return raw
}

func normalizeTToken(raw string) string {
	raw = normalizeNumberToken(raw)
	parts := strings.SplitN(raw, ".", 2)
	intPart := parts[0]
	frac := ""
	if len(parts) == 2 {
		frac = parts[1]
	}
	if frac == "" {
		frac = "0"
	}
	if len(intPart) >= 2 {
		return intPart + "." + frac
	}
	if len(frac) > 1 {
		frac = frac[:1]
	}
	return intPart + "." + frac
}

func atoi(v string) int {
	n := 0
	for _, r := range v {
		if r < '0' || r > '9' {
			continue
		}
		n = n*10 + int(r-'0')
	}
	return n
}

// compressArticle shortens an over-long article by progressively dropping/abbreviating
// segments. It returns the (possibly shortened) segment list; the caller re-joins it.
func compressArticle(segs []namedSeg) []namedSeg {
	if len(segs) == 0 {
		return segs
	}
	fits := func() bool {
		return len([]rune(strings.Join(namedSegsValues(segs), "-"))) <= 80
	}
	for i, s := range segs {
		segs[i].value = strings.ReplaceAll(s.value, "GbE", "G")
	}
	if fits() {
		return segs
	}

	// 1) remove PSU
	if i := findSegGroup(segs, "PSU"); i >= 0 {
		segs = append(segs[:i], segs[i+1:]...)
		if fits() {
			return segs
		}
	}

	// 2) compress NET/HBA/HCA
	if i := findSegGroup(segs, "NET"); i >= 0 {
		segs[i].value = compressNetSegment(segs[i].value)
		if fits() {
			return segs
		}
	}

	// 3) compress DISK
	if i := findSegGroup(segs, "DISK"); i >= 0 {
		segs[i].value = compressDiskSegment(segs[i].value)
		if fits() {
			return segs
		}
	}

	// 4) compress GPU to vendor only (GPU_NV)
	if i := findSegGroup(segs, "GPU"); i >= 0 {
		segs[i].value = compressGPUSegment(segs[i].value)
	}
	return segs
}

func compressNetSegment(seg string) string {
	if seg == "" {
		return seg
	}
	parts := strings.Split(seg, "+")
	out := make([]string, 0, len(parts))
	for _, p := range parts {
		p = strings.TrimSpace(p)
		if p == "" {
			continue
		}
		qty := "1"
		profile := p
		if x := strings.SplitN(p, "x", 2); len(x) == 2 {
			qty = x[0]
			profile = x[1]
		}
		upper := strings.ToUpper(profile)
		label := "NIC"
		if strings.Contains(upper, "FC") {
			label = "HBA"
		} else if strings.Contains(upper, "HCA") || strings.Contains(upper, "IB") {
			label = "HCA"
		}
		out = append(out, fmt.Sprintf("%sx%s", qty, label))
	}
	if len(out) == 0 {
		return seg
	}
	sort.Strings(out)
	return strings.Join(out, "+")
}

func compressDiskSegment(seg string) string {
	if seg == "" {
		return seg
	}
	parts := strings.Split(seg, "+")
	out := make([]string, 0, len(parts))
	for _, p := range parts {
		p = strings.TrimSpace(p)
		if p == "" {
			continue
		}
		qty := "1"
		spec := p
		if x := strings.SplitN(p, "x", 2); len(x) == 2 {
			qty = x[0]
			spec = x[1]
		}
		upper := strings.ToUpper(spec)
		label := "DSK"
		for _, t := range []string{"M2", "NV", "SAS", "SAT", "SSD", "HDD", "EDS", "HHH"} {
			if strings.Contains(upper, t) {
				label = t
				break
			}
		}
		out = append(out, fmt.Sprintf("%sx%s", qty, label))
	}
	if len(out) == 0 {
		return seg
	}
	sort.Strings(out)
	return strings.Join(out, "+")
}

func compressGPUSegment(seg string) string {
	if seg == "" {
		return seg
	}
	parts := strings.Split(seg, "+")
	out := make([]string, 0, len(parts))
	for _, p := range parts {
		p = strings.TrimSpace(p)
		if p == "" {
			continue
		}
		qty := "1"
		if x := strings.SplitN(p, "x", 2); len(x) == 2 {
			qty = x[0]
		}
		out = append(out, fmt.Sprintf("%sxGPU_NV", qty))
	}
	if len(out) == 0 {
		return seg
	}
	sort.Strings(out)
	return strings.Join(out, "+")
}