From d4a1ce48e436c42f87a42c8d0817bea4d120a5fa Mon Sep 17 00:00:00 2001
From: Danila Kamaev <github@kamaeff.com>
Date: Mon, 31 Aug 2026 18:19:17 +0400
Subject: [PATCH] thumbex: collabora example (#1617)

---
 bin/thumbs/README.md    |   1 +
 bin/thumbs/collabora.py | 125 ++++++++++++++++++++++++++++++++++++++++
 2 files changed, 126 insertions(+)
 create mode 100644 bin/thumbs/collabora.py

diff --git a/bin/thumbs/README.md b/bin/thumbs/README.md
index 0c2018c6a..ffb0f6c68 100644
--- a/bin/thumbs/README.md
+++ b/bin/thumbs/README.md
@@ -47,6 +47,7 @@ copyparty performs buffered reading, so expect multiple `read()` calls, respect
 ## examples
 
 * [randomcolor](https://github.com/9001/copyparty/blob/hovudstraum/bin/thumbs/randomcolor.py) generates a random .gif image
+* [collabora](https://github.com/9001/copyparty/blob/hovudstraum/bin/thumbs/collabora.py) uses external self-hosted service (collabora online) to generate previews for various office documents, pdf and fb2 books, html, markdown, text files and some image formats; collabora is also supported as office editor in copyparty, see [wopi](https://github.com/9001/copyparty#wopi-server)
 
 
 ## some other known plugins seen on the internets
diff --git a/bin/thumbs/collabora.py b/bin/thumbs/collabora.py
new file mode 100644
index 000000000..d215b6a9b
--- /dev/null
+++ b/bin/thumbs/collabora.py
@@ -0,0 +1,125 @@
+"""
+CAUTION!
+looks like collabora spawns separate jailed libreoffice instance to do the conversion
+this may eat up too much ram in folders with many docs or with --th-pregen enabled
+
+consider limiting resources available for collabora
+e.g. for me these lines in docker-compose.yml helped alot:
+    deploy:
+      resources:
+        limits:
+          cpus: '2'
+          memory: '2G'
+/CAUTION!
+
+if you use collabora office editor with copyparty, you already have a thumbnailer for office formats:
+https://sdk.collaboraonline.com/docs/conversion_api.html
+the 'convert-to' api accepts files in multipart/form-data format
+
+just add this extractor to copyparty with an argument like this:
+  --th-extract=docx,xlsx,pptx,odt,ods,pdf,fb2=~/dev/copyparty/bin/thumbs/collabora.py
+and set COLLABORA_INSTANCE_URL below to your collabora instace url
+
+full list of supported file types:
+https://sdk.collaboraonline.com/docs/conversion_api.html#supported-input-formats
+
+this is also an example of deferred IO:
+no files opened and no network requests made if copyparty doesn't like the file type
+"""
+
+import os
+import string
+import random
+import mimetypes
+from urllib.error import HTTPError
+from urllib.request import Request, urlopen
+
+COLLABORA_INSTANCE_URL = "http://collabora:9980"
+BOUNDARY_CHARS = string.ascii_letters + string.digits
+
+
+def main(abspath, **kwargs):
+    boundary = "CopypartyFormData" + "".join(random.choices(BOUNDARY_CHARS, k=53))
+    start, end = make_headers(abspath, boundary)
+    size = len(start) + os.path.getsize(abspath) + len(end)
+    request = Request(
+        COLLABORA_INSTANCE_URL + "/cool/convert-to/png",
+        method="POST",
+        data=iterate(start, abspath, end),
+        headers={
+            "Content-Type": f"multipart/form-data; boundary={boundary}",
+            "Content-Length": str(size),
+        },
+    )
+
+    file_like = Wrapper(request)
+    offset, whence, length = 0, 0, None
+
+    return "png", file_like, offset, whence, length
+
+
+def make_headers(abspath, boundary):
+    """stuff to wrap file contents by for multipart"""
+
+    fname = "file." + abspath.rsplit(".", 1)[-1]
+    ftype = mimetypes.guess_file_type(fname)[0] or "application/octet-stream"
+    start = (
+        f"--{boundary}\r\n"
+        f'Content-Disposition: form-data; name="data"; filename="{fname}"\r\n'
+        f"Content-Type: {ftype}\r\n\r\n"
+    )
+    end = f"\r\n--{boundary}--"
+    return bytes(start, "ascii"), bytes(end, "ascii")
+
+
+def iterate(start, fpath, end):
+    """docs may be big so read and send them in chunks"""
+
+    yield start
+    with open(fpath, "rb") as f:
+        while True:
+            b = f.read(32768)
+            if not b:
+                break
+            yield b
+    yield end
+
+
+class Wrapper:
+    """
+    This file-like wrapper defers expensive work until first .seek() call
+    Useful  to not waste resources in case copyparty decides
+    to throw our thumbnail away based on returned file extension
+    """
+
+    def __init__(self, request):
+        self.request = request
+        self.response = None
+
+    def seek(self, offset, whence=0):
+        if self.response:
+            return 0
+
+        try:
+            # the heavy part:
+            # - reading probably big file
+            # - sending it over the network
+            # - making collabora do some work
+            self.response = urlopen(self.request, timeout=10)
+        except HTTPError as e:
+            e.close()
+            raise e
+        except Exception as e:
+            self.close()
+            raise e
+
+        # actual response is not seekable
+        # just return 0
+        return 0
+
+    def read(self, amount):
+        return self.response.read(amount)
+
+    def close(self):
+        if self.response:
+            self.response.close()