This project began as a port of the html_to_plaintext logic from github.com/premailer/premailer and applies the same basic set of rules for generating a text/plain copy of an email, given the text/html version.
go get github.com/mailproto/textplainmyHTML := `<html><body>Hello World</body></html>`
myPlaintext, err := textplain.Convert(myHTML)
if err != nil {
// ErrBodyNotFound if the document has no body element
}ConvertReader does the same for an io.Reader, such as a MIME part, and also returns any error from reading it.
ConvertNode takes a document already parsed with html.Parse, for a caller that holds the tree and would otherwise render it to a string only for it to be parsed again. It does not modify the tree.
Input is read as UTF-8. For a part in another charset, such as ISO-8859-1, decode it first with golang.org/x/net/html/charset:
utf8Reader, err := charset.NewReader(part, part.Header.Get("Content-Type"))
if err != nil {
// unknown charset
}
myPlaintext, err := textplain.ConvertReader(utf8Reader)Output wraps at DefaultLineLength, which is 65. The word wrapping is exported for use on its own. It counts columns rather than bytes, with Chinese, Japanese and Korean characters taking two, and breaks between those characters as well as at spaces:
wrapped := textplain.WordWrap("hello world, here is some text", 15)Pass a line length of zero or less to skip wrapping entirely.
Convert takes any number of options after the document.
myPlaintext, err := textplain.Convert(myHTML,
textplain.WithLinks(textplain.LinksFootnotes),
textplain.WithPlainHeadings(),
)| option | default |
|---|---|
WithLineLength(columns) |
DefaultLineLength; zero or less does not wrap |
WithBullet(prefix) |
"* " on unordered list items |
WithOrderedSuffix(suffix) |
". " after the number on ordered list items |
WithLinks(style) |
LinksInline |
WithPlainHeadings() |
off, so headings are drawn with rule characters |
WithMarkdown() |
off; renders CommonMark instead of plain text |
WithHiddenContent() |
off, so content hidden by inline styles or attributes is dropped |
Later options win, so a caller can layer its own on top of a shared set.
Hidden content is judged from inline styles, so a layout can occasionally convert to an empty string. Callers that need text can retry with WithHiddenContent().
WithLinks takes one of three styles:
| style | <a href="https://example.com">Docs</a> becomes |
|---|---|
LinksInline |
Docs ( https://example.com ) |
LinksOmitted |
Docs |
LinksFootnotes |
Docs [1], with the targets listed under the body |
LinksFootnotes numbers each target in the order it appears:
Read the Docs [1] or the changelog [2].
[1] https://example.com
[2] https://example.com/changelog
A table whose first row is all header cells is laid out in columns, with the header ruled off:
Item Qty Price
------------ --- ------
Widget 2 $10.00
Any other table is treated as layout and its cells run on as text. So is a table with nested tables, cells spanning rows, cells holding more than a line, or columns too wide for the line length.
WithMarkdown renders the same content as CommonMark: # headings, **bold**, *italic* and ~~struck~~, code spans and fences,  images (just the alt for inline data: images), and [text](url) links, <url> where the text is the address, or reference links with LinksFootnotes. Text that would otherwise read as markup is escaped. Data tables become GFM tables; layout tables are flattened as in plain text, with each row ending in a hard break.
Markdown output is never wrapped, so WithLineLength and WithPlainHeadings have no effect.