129614 - low space in target directory: no error dialog, incomplete download

Status: VERIFIED FIXED

(Core Graveyard :: File Handling, defect)

Description

[sairuh (rarely reading bugmail)]

found while verifying scenarios for bug 27609, on linux rh7.2 using
2002.03.06.07 comm bits.

1. unlike bug 129604, /tmp is no longer full. instead, i fill up another
partition where i'll be saving the file --in my case /export/builds/foopy

2. click to download the following 35M file:
ftp://ftp.mozilla.org/pub/mozilla/nightly/2002-03-07-08-trunk/mozilla-source.tar.gz

3. in the file picker, select /export/builds/foopy as the target location.

results: no error dialog appears. the download progress dialog showed
completion, too. however, when looking at the contents of /export/builds/foopy,
the file there was incomplete --definitely less than the expected 35M. moreover,
the salted file persisted in /tmp even after quitting.

later on, i'll try this out by saving to a floppy disk.

btw, not a problem on mac 10.1.3 when i tried saving to a Zip disk which didn't
have enough space.

will test win2k shortly...

Comment 1

[sairuh (rarely reading bugmail)]

this is also a problem on win2k, using 2002.03.06.06 comm verif bits. i tried
saving the same 35M file to a floppy and a ramdisk --both lacking the space.
there was no error dlg, and the download progress dlg appeared done.

difference in results: no file is saved at all, and no lingering temp file in
C:\WINNT\Temp...

Comment 2

[Bill Law]

I think this could be due to insufficient error handling when moving/copying the
downloaded file from temp to final destination.  There may already be a bug on that.

Comment 3

[Peter Trudelle]

We really should ensure adequate space exists on the destination at the time it
is chosen, and preferably copy the bits directly there.  Using an intermediate
file just introduces lots more failure cases, and takes extra time and disk space.  

Comment 4

[Peter Trudelle]

nsbeta1+ per Nav triage team, should at least notify user.

Comment 5

[Peter Trudelle]

nav2

Comment 6

[Matthias Versen [:Matti]]

sairuh: isn't this a duipe of bug 115201 ?

Comment 7

[selmer (gone)]

Please update this bug with an [adt1] - [adt3] impact rating (or take it off the
list if it doesn't even rate adt3.)  Thanks!

Comment 8

[Bill Law]

Changing from nsbeta1+ back to nsbeta1 (nominated only), so the triage team can
decide.

Comment 9

[Samir Gehani]

Nav triage team: nsbeta1+/adt2

Comment 10

[John Villalovos]

I have noticed this with 1.0rc1.  If the disk is completly full before the
download the download appears to be successful but there is no file on the disk.

This was when I was saving to a Samba share that was completely full.  This was
quite confusing for awhile because I could not figure out why the file was not
there.

Comment 11

[Bill Law]

Attachment: fix, v1.0

This seems to fix most problems, still needs more testing before it's done.  I
have to upload it here so I can do other work on these files.

Comment 12

[Bill Law]

Attachment: patch v2.0

Revised patch including changes to download manager.

Ready for review...

Comment 13

[Bill Law]

Attachment: Explanation of changes in patch

Please consult this for explanation of the fix.

Comment 14

[Boris Zbarsky [:bzbarsky]]

So... the alert we put up in the "really rare case" has no parent.  Does this
not mean that it is non-modal?  Should we not use mWindowContext as the parent? 
(QI to nsIInterfaceRequestor and GetInterface(nsIDOMWindowInternal) as
nsHelperAppDlg.js does).

Other than that, this looks great. r=bzbarsky if you address the alert-parent
issue.

Comment 15

[Blake Ross]

Comment on attachment 82747 [details] [diff] [review]
patch v2.0

sr=blake
Thanks for those comments. That really facilitates review.

A few minor nits:

// Exception mans file doesn't exist; launch is not allowed.

Please fix that typo.

And please don't wrap the condition of that |if| with spaces in
nsDownloadManager.cpp.

Comment 16

[Bill Law]

Attachment: patch v3.0, incorporating Boris's suggestion

I added this line to get the parent from mWindowContext:

+  nsCOMPtr<nsIDOMWindow> parent(do_GetInterface(mWindowContext));

This required a bit more work because SendStatusChange wasn't a member function
and to get at mWindowContext I had to make it one).  I also tweaked some of the
error handling code to make sure Cancel() is called whenever an error occurs
(that was necessary to get the download manager to mark failed downloads as
such rather than "completed").

Comment 17

[Boris Zbarsky [:bzbarsky]]

Comment on attachment 83824 [details] [diff] [review]
patch v3.0, incorporating Boris's suggestion

r=bzbarsky

looks good.  Don't forget to check in Mac build changes if needed when you land
this...

Comment 18

[Bill Law]

fixed

Comment 19

[Bill Law]

no mac build changes required

fixed

Comment 20

[Judson Valeski]

Comment on attachment 83824 [details] [diff] [review]
patch v3.0, incorporating Boris's suggestion

assuming carry over of sr. please slap me if I'm wrong. also approving patch.

nice comments!

Comment 21

[Judson Valeski]

please checkin to the 1.0.1 branch ASAP. once there please remove the
mozilla1.0.1+ keyword and add the fixed1.0.1 keyword.

Comment 22

[Jaime Rodriguez, Jr.]

Mass removing adt1.0.0, and adding adt1.0.1 because, we are now on 1.0.1.

Comment 23

[scottputterman]

sairuh, could you verify this on the trunk?

Comment 24

[sairuh (rarely reading bugmail)]

couldn't get to this till now (was on vacation during comment 23).

on win2k (2002.06.13.08 commercial trunk), this works fine now. tested both the
context menu ("save link target") as well as via helper app dlg (single-click
downloadable link).

on mac 10.1.5 (2002.06.13.08 commercial trunk), this also works (same tests as
with win2k). the only hitch here is that it took a while for the error dlg to
actually appear. ie, for the "save link target" test, 16% was downloaded before
an error dlg appeared. and for the helper app dlg, progress went as far as 100%
before the error appeared.

the Mac OS X behavior is not very friendly, but i'd say the particular issue of
getting no dlg at all is fixed on the trunk.

Comment 25

[sairuh (rarely reading bugmail)]

retested with today's mac os x trunk build: didn't seem to take as long when
using "save link target as" --so filed bug 152392 for the "waiting till 100%..."
issue.

Comment 26

[scottputterman]

adding adt1.0.1+.  Please get drivers approval again before checking into the
branch.

Comment 27

[Boris Zbarsky [:bzbarsky]]

If/when this lands on the branch, bug 147142 should land as well.

Comment 28

[Jaime Rodriguez, Jr.]

law: was this patch checked into the branch? if yes, pls replace "mozilla1.0.1+"
with the "fixed1.0.1" keyword.

Comment 29

[Jaime Rodriguez, Jr.]

adt1.0.1- per the ADT. let's get it in the next release.

Comment 30

[Randell Jesup [:jesup] (needinfo me)]

Retargeting since no one is pushed this for 1.0.1 (no one re-asked for approval).