commit 60e6b177730ef068415639bf48471431dc69870b
parent bf40c1f2315b429d1d058b19e306a729f67de515
Author: Michael Neuper <michael@michaelneuper.com>
Date: Sun, 25 Oct 2026 22:07:27 +0200
resolve relative links and images in the README
Paths resolve from the repository root, like on GitHub. Links to text
files go to their file page and links to directories to their listing.
Images and other binary files are copied to raw/ and linked there.
Diffstat:
| M | README.md | | | 6 | ++++-- |
| M | stagit.c | | | 173 | +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++-- |
2 files changed, 174 insertions(+), 5 deletions(-)
diff --git a/README.md b/README.md
@@ -90,5 +90,7 @@ Merge rather than rebase, so `main` never needs a force push. Conflicts in
existing commit page, so incremental output would go stale after a force
push or a description change. Netlify only uploads files whose contents
changed, so redeploys stay small.
-- Relative image links in READMEs don't work, because stagit only publishes
- HTML views of files, not the files themselves.
+- Relative links and images in a README resolve from the repo root, like on
+ GitHub. Links to text files go to their file page and links to directories
+ to their listing. Images and other binary files are copied to
+ `raw/<path>` and linked there.
diff --git a/stagit.c b/stagit.c
@@ -1,6 +1,7 @@
#include <sys/stat.h>
#include <sys/types.h>
+#include <ctype.h>
#include <err.h>
#include <errno.h>
#include <libgen.h>
@@ -984,13 +985,172 @@ writemarkdown(const MD_CHAR *s, MD_SIZE len, void *fp)
fwrite(s, 1, len, fp);
}
+/* If url is a path in the repository, resolved from the root the way GitHub
+ * does for a README, write the URL of the page for it instead: the listing
+ * for a directory and the file page for a text file. Images and other binary
+ * files are copied to raw/ and linked there. Returns 0 for anything else,
+ * such as an absolute URL. */
+int
+writereadmeurl(FILE *fp, const char *url, size_t len, int img)
+{
+ git_object *obj = NULL;
+ char path[PATH_MAX], spec[PATH_MAX], out[PATH_MAX], fpath[PATH_MAX];
+ char tmp[PATH_MAX], hex[3] = "", *d;
+ const char *s;
+ size_t i, j;
+ FILE *rawfp;
+ int r;
+
+ /* absolute URLs have a scheme or start with // */
+ for (i = 0; i < len && !strchr("/?#", url[i]); i++)
+ if (url[i] == ':')
+ return 0;
+ if (!len || url[0] == '#' || url[0] == '?' ||
+ (len > 1 && url[0] == '/' && url[1] == '/'))
+ return 0;
+
+ /* decode the path, without a query string or fragment */
+ for (i = 0, j = 0; i < len && url[i] != '?' && url[i] != '#'; i++) {
+ if (j + 1 >= sizeof(path))
+ return 0;
+ if (url[i] == '%' && i + 2 < len &&
+ isxdigit((unsigned char)url[i + 1]) &&
+ isxdigit((unsigned char)url[i + 2])) {
+ hex[0] = url[i + 1];
+ hex[1] = url[i + 2];
+ path[j++] = strtol(hex, NULL, 16);
+ i += 2;
+ } else if (i + 5 <= len && !strncmp(&url[i], "&", 5)) {
+ path[j++] = '&';
+ i += 4;
+ } else {
+ path[j++] = url[i];
+ }
+ }
+ while (j > 0 && path[j - 1] == '/')
+ j--;
+ path[j] = '\0';
+
+ /* "/dir/file" and "./dir/file" mean "dir/file" */
+ for (s = path; *s == '/' || !strncmp(s, "./", 2); s += (*s == '/') ? 1 : 2)
+ ;
+ /* never write outside raw/ */
+ j = strlen(s);
+ if (!strcmp(s, "..") || !strncmp(s, "../", 3) || strstr(s, "/../") ||
+ (j >= 3 && !strcmp(s + j - 3, "/..")))
+ return 0;
+
+ /* the repository root */
+ if (!*s || !strcmp(s, ".")) {
+ if (img)
+ return 0;
+ fputs("files.html", fp);
+ return 1;
+ }
+
+ r = snprintf(spec, sizeof(spec), "HEAD:%s", s);
+ if (r < 0 || (size_t)r >= sizeof(spec) ||
+ git_revparse_single(&obj, repo, spec)) {
+ git_object_free(obj);
+ return 0;
+ }
+ outpath(out, sizeof(out), s, strlen(s));
+
+ switch (git_object_type(obj)) {
+ case GIT_OBJ_TREE:
+ if (img)
+ break;
+ fputs("tree/", fp);
+ percentencode(fp, out, strlen(out));
+ fputs(".html", fp);
+ git_object_free(obj);
+ return 1;
+ case GIT_OBJ_BLOB:
+ if (!img && !git_blob_is_binary((git_blob *)obj)) {
+ fputs("file/", fp);
+ percentencode(fp, out, strlen(out));
+ fputs(".html", fp);
+ git_object_free(obj);
+ return 1;
+ }
+
+ r = snprintf(fpath, sizeof(fpath), "raw/%s", out);
+ if (r < 0 || (size_t)r >= sizeof(fpath))
+ errx(1, "path truncated: 'raw/%s'", out);
+ if (strlcpy(tmp, fpath, sizeof(tmp)) >= sizeof(tmp))
+ errx(1, "path truncated: '%s'", fpath);
+ if (!(d = dirname(tmp)))
+ err(1, "dirname");
+ if (mkdirp(d))
+ break;
+
+ rawfp = efopen(fpath, "w");
+ fwrite(git_blob_rawcontent((git_blob *)obj), 1,
+ git_blob_rawsize((git_blob *)obj), rawfp);
+ checkfileerror(rawfp, fpath, 'w');
+ fclose(rawfp);
+
+ fputs("raw/", fp);
+ percentencode(fp, out, strlen(out));
+ git_object_free(obj);
+ return 1;
+ default:
+ break;
+ }
+
+ git_object_free(obj);
+ return 0;
+}
+
+/* rendered README, with links and image sources that are paths in the
+ * repository pointing at the pages or copies for them */
+void
+writereadmehtml(FILE *fp, const char *s)
+{
+ const char *tag, *end, *attr, *a, *url, *q;
+ size_t n;
+ int img;
+
+ while ((tag = strchr(s, '<')) && (end = strchr(tag, '>'))) {
+ img = !strncmp(tag, "<img", 4) && isspace((unsigned char)tag[4]);
+ if (img)
+ attr = "src=";
+ else if (!strncmp(tag, "<a", 2) && isspace((unsigned char)tag[2]))
+ attr = "href=";
+ else
+ attr = NULL;
+
+ n = attr ? strlen(attr) : 0;
+ for (a = tag + 1; attr && a < end; a++) {
+ if (isspace((unsigned char)a[-1]) && !strncmp(a, attr, n) &&
+ (a[n] == '"' || a[n] == '\''))
+ break;
+ }
+ if (!attr || a >= end) {
+ fwrite(s, 1, end - s, fp);
+ s = end;
+ continue;
+ }
+
+ url = a + n + 1;
+ if (!(q = strchr(url, a[n])))
+ break;
+ fwrite(s, 1, url - s, fp);
+ if (!writereadmeurl(fp, url, q - url, img))
+ fwrite(url, 1, q - url, fp);
+ s = q;
+ }
+ fputs(s, fp);
+}
+
void
writereadme(FILE *fp)
{
git_object *obj = NULL;
- char spec[PATH_MAX];
+ char spec[PATH_MAX], *html = NULL;
const char *s, *ext;
- size_t len;
+ size_t len, htmllen;
+ FILE *htmlfp;
snprintf(spec, sizeof(spec), "HEAD:%s", readme);
if (git_revparse_single(&obj, repo, spec))
@@ -1000,8 +1160,15 @@ writereadme(FILE *fp)
fputs("<div id=\"readme\">\n", fp);
if ((ext = strrchr(readme, '.')) && !strcmp(ext, ".md")) {
- if (md_html(s, len, writemarkdown, fp, MD_DIALECT_GITHUB, 0))
+ /* render to memory first, to rewrite image sources */
+ if (!(htmlfp = open_memstream(&html, &htmllen)))
+ err(1, "open_memstream");
+ if (md_html(s, len, writemarkdown, htmlfp, MD_DIALECT_GITHUB, 0))
errx(1, "md_html: '%s'", readme);
+ if (fclose(htmlfp))
+ err(1, "fclose: '%s'", readme);
+ writereadmehtml(fp, html);
+ free(html);
} else {
fputs("<pre>", fp);
xmlencode(fp, s, len);
