diff options
| author | Dax Raad <[email protected]> | 2025-05-18 22:30:41 -0400 |
|---|---|---|
| committer | Dax Raad <[email protected]> | 2025-05-26 12:40:17 -0400 |
| commit | 99af6146d5def31c59993636d60eb75a483a283b (patch) | |
| tree | eb57ed227c15cf9be54bc9f231c83895d812f881 /pkg/tui/styles/background.go | |
| parent | 020e0ca039287b73fa33041fbd1bb214e6ccb396 (diff) | |
| download | opencode-99af6146d5def31c59993636d60eb75a483a283b.tar.gz opencode-99af6146d5def31c59993636d60eb75a483a283b.zip | |
openapi
Diffstat (limited to 'pkg/tui/styles/background.go')
| -rw-r--r-- | pkg/tui/styles/background.go | 123 |
1 files changed, 123 insertions, 0 deletions
diff --git a/pkg/tui/styles/background.go b/pkg/tui/styles/background.go new file mode 100644 index 000000000..2fbb34efb --- /dev/null +++ b/pkg/tui/styles/background.go @@ -0,0 +1,123 @@ +package styles + +import ( + "fmt" + "regexp" + "strings" + + "github.com/charmbracelet/lipgloss" +) + +var ansiEscape = regexp.MustCompile("\x1b\\[[0-9;]*m") + +func getColorRGB(c lipgloss.TerminalColor) (uint8, uint8, uint8) { + r, g, b, a := c.RGBA() + + // Un-premultiply alpha if needed + if a > 0 && a < 0xffff { + r = (r * 0xffff) / a + g = (g * 0xffff) / a + b = (b * 0xffff) / a + } + + // Convert from 16-bit to 8-bit color + return uint8(r >> 8), uint8(g >> 8), uint8(b >> 8) +} + +// ForceReplaceBackgroundWithLipgloss replaces any ANSI background color codes +// in `input` with a single 24‑bit background (48;2;R;G;B). +func ForceReplaceBackgroundWithLipgloss(input string, newBgColor lipgloss.TerminalColor) string { + // Precompute our new-bg sequence once + r, g, b := getColorRGB(newBgColor) + newBg := fmt.Sprintf("48;2;%d;%d;%d", r, g, b) + + return ansiEscape.ReplaceAllStringFunc(input, func(seq string) string { + const ( + escPrefixLen = 2 // "\x1b[" + escSuffixLen = 1 // "m" + ) + + raw := seq + start := escPrefixLen + end := len(raw) - escSuffixLen + + var sb strings.Builder + // reserve enough space: original content minus bg codes + our newBg + sb.Grow((end - start) + len(newBg) + 2) + + // scan from start..end, token by token + for i := start; i < end; { + // find the next ';' or end + j := i + for j < end && raw[j] != ';' { + j++ + } + token := raw[i:j] + + // fast‑path: skip "48;5;N" or "48;2;R;G;B" + if len(token) == 2 && token[0] == '4' && token[1] == '8' { + k := j + 1 + if k < end { + // find next token + l := k + for l < end && raw[l] != ';' { + l++ + } + next := raw[k:l] + if next == "5" { + // skip "48;5;N" + m := l + 1 + for m < end && raw[m] != ';' { + m++ + } + i = m + 1 + continue + } else if next == "2" { + // skip "48;2;R;G;B" + m := l + 1 + for count := 0; count < 3 && m < end; count++ { + for m < end && raw[m] != ';' { + m++ + } + m++ + } + i = m + continue + } + } + } + + // decide whether to keep this token + // manually parse ASCII digits to int + isNum := true + val := 0 + for p := i; p < j; p++ { + c := raw[p] + if c < '0' || c > '9' { + isNum = false + break + } + val = val*10 + int(c-'0') + } + keep := !isNum || + ((val < 40 || val > 47) && (val < 100 || val > 107) && val != 49) + + if keep { + if sb.Len() > 0 { + sb.WriteByte(';') + } + sb.WriteString(token) + } + // advance past this token (and the semicolon) + i = j + 1 + } + + // append our new background + if sb.Len() > 0 { + sb.WriteByte(';') + } + sb.WriteString(newBg) + + return "\x1b[" + sb.String() + "m" + }) +} |
