On the fly bundling of multiple GZipStreams into zip.

[simonmacarthur]

Hi,
After some advice before I go off and build something from scratch.

My solution stores a number of PDF files uploaded by users up onto Azure Blob storage. These files are streamed through * the GZipStream class during upload, with the server-side file stored in compressed form. During download of such a pdf, the appropriate headers response headers are set and the browser automatically decompresses and views the files as expected.

My next task is to allow users to select multiple files and download them as a singular zip file.

Clearly I could build temporary zips easily enough, store the file, then stream the result to the browser, however it seems I have enough information to build a "virtual zip" using a combination of on-disk ".gz"s and stored data captured during initial upload (CRC, original file size, compressed file size, name), and stream this directly to the browser on-the-fly.

I'd need to calculate the total output size of the final zip stream in advance to produce the correct HTML response header length, but then stream the relevant zip local file headers, my persisted gz streams, and finally the central file headers to the output stream....

Thoughts or pointers gratefully received,

Si

[piksel]

The new "precompresed entity" support (from PR #683) should help with this. I'm not sure about calculating the size though...

[simonmacarthur]

Thanks @piksel

Playing around with PutNextPassthroughEntry I think I've come up with a workable (if slightly inelegant solution).

Firstly I build a ZipOutputStream on top of a MemoryStream. For each item file I intend to be contained in the final output stream I create a zero length ZipEntry. After calling the finish method the memory stream's position is the "overhead" for the zip header / footer / catalog entries. Adding this overhead number to the sum of all my compressed files gives the correct final zip size.

A little clunky (and I suspect with some bugs if used for 64bit sized files), however should do for now.