gzip files expanded when downloaded via "save link as" go to http://ftp.mozilla.org/pub/mozilla/nightly/latest/, and right click on a .tar.gz file. select "save link as" and notice that the downloaded file has been decompressed. this happens because the code invoking the download has not set nsIHttpChannel::SetApplyConversion(PR_FALSE) as it should when downloading files. see nsStreamTransfer.cpp:152 for an example of how this works.
->default owner, but punt as needed.
I also noticed that when I use a proxy that the downloaded file is both ungzipped and TRUNCATED. Is there a bug on that?
Japanese users reported this bug reproduced MacOS 9.2(2001122408 ) and MacOSX(0.9.7). Platform and OS should be All/All.
This sounds like a bug in WebBrowserPersist itself... Adding dataloss since the file is truncated.
The xpfe should be specifying PERSIST_FLAGS_NO_CONVERSION as a flag to the webbrowserpersist object. See bug 108268
nominating since this should be a simple fix for a dataloss bug.... Just add a flag setting at http://lxr.mozilla.org/seamonkey/source/embedding/components/ui/progressDlg/nsPr ogressDlg.js#382 (right before the try/catch). + webBrowserPersist.persistFlags |= + Components.interfaces.nsIWebBrowserPersist.PERSIST_FLAGS_NO_CONVERSION; (if I recall the syntax for getting to interface constants correctly). I'd attach a patch if I had a tree...
*** Bug 118041 has been marked as a duplicate of this bug. ***
*** Bug 118320 has been marked as a duplicate of this bug. ***
16 years ago
That specifies "no conversion" all the time, right? I think there might be a complication when saving a web page served up by a server that is doing content-encoding. There has always been some lurking bugs about that (or similar) situation and we should try to squash that now. I'll have to go track down the discussion. Maybe we just turn off encoding for non-text types? Or only turn it off for application/zip, etc.?
One thing is that Apache will serve up gzipped files as Content-Type: application/x-gzip (or application/gzip?) Content-Encoding: gzip which is, of course, wrong.... We should try to handle that case as we fix this. :(
bz: that case is already handled internally by necko. all we need to do is ensure that necko is instructed to not apply content conversions. see the top of nsHttpChannel::ProcessNormal for code that handles apache.
Darin, would the "no conversion ever" patch do the right thing then? If not, under what conditions should the "no conversion" flag be set such that gzip-encoded HTML is gunzipped but regular gzip files are not?
if that's the behavior you want then you need to check for Content-Type and Content-Encoding headers. telling necko to not apply content conversions means that necko will simply ignore the Content-Encoding header. by default necko honors a Content-Encoding header. you can override this behavior in your OnStartRequest handler.
Darin, How does this interact with the cache? Is the content in the cache unzipped, or compressed? If we ask for the data from the cache with no content encoding, might we still get it uncompressed if it is fetched from the cache?
the content in the cache is decoded per any transfer encodings specified by a Transfer-Encoding header and encoded per any content encodings specified by a Content-Encoding header. this means that if a server sends a document with Content-Encoding: gzip, the document in the cache will be compressed. you have actually described a bug that existed in the old days... the compressed state of a saved file used to depend on how it was originally fetched. now, we avoid any problems like that by writing the data to the cache before (possibly) decoding an encoded document.
What is the relation between this and bug 35956 ? Would it help people working on this to compare?
The checkin for bug 110135 made PERSIST_FLAGS_NO_CONVERSION a default persist flag so it would be worth retesting now. Even so, it's best for the client (xpfe) to be explicit and set the flags it wants to use unless it doesn't care how the persist object saves something.
Fixed. Patch w/analysis in bug 120174.
Works ok in 0.9.8. Anyone wants to mark "VERIFIED" or to close? BTW, I do understand that "Content-encoding: gzip" is wrong. Do you guys know what is right? ": data"?
verified that we now never unzip, even for files for which we should. I'll address that in another bug...