From 2463d4b855edfc5b8f7502c5ba40fc831f955492 Mon Sep 17 00:00:00 2001 From: Joey Hess Date: Tue, 15 May 2018 11:54:02 -0400 Subject: [PATCH] response --- ..._b129b1642f5d21a5b933c8c51807b33c._comment | 22 +++++++++++++++++++ ..._b7c2abf63c9f350402b3176e028bda03._comment | 21 ++++++++++++++++++ 2 files changed, 43 insertions(+) create mode 100644 doc/bugs/Metadata_views_in_v6_repo_upgraded_from_direct_mode_act_strangely/comment_2_b129b1642f5d21a5b933c8c51807b33c._comment create mode 100644 doc/bugs/annex_view_barfs__fatal__58___Unable_to_add___40__null__41___to_database/comment_4_b7c2abf63c9f350402b3176e028bda03._comment diff --git a/doc/bugs/Metadata_views_in_v6_repo_upgraded_from_direct_mode_act_strangely/comment_2_b129b1642f5d21a5b933c8c51807b33c._comment b/doc/bugs/Metadata_views_in_v6_repo_upgraded_from_direct_mode_act_strangely/comment_2_b129b1642f5d21a5b933c8c51807b33c._comment new file mode 100644 index 0000000000..e66d61afc6 --- /dev/null +++ b/doc/bugs/Metadata_views_in_v6_repo_upgraded_from_direct_mode_act_strangely/comment_2_b129b1642f5d21a5b933c8c51807b33c._comment @@ -0,0 +1,22 @@ +[[!comment format=mdwn + username="joey" + subject="""comment 2""" + date="2018-05-15T15:37:57Z" + content=""" +This has nothing to do with direct mode or v6, or the filesystem. +It is behaving more or less as intended. + +Filenames in views are always mangled to make them guaranteed to be +unique, since two files with base name "foo" could be collected into the +same view directory. This is done by embedding the directory structure +inside the filename as you show. + +In the specific case of "/=*" that is not necessary since the view +is replicating the full directory structure. But most views do need +the mangling, and I feel it's better to always do the mangling than try to +work out when it's not needed. + +And "/=*" is a mostly not useful edge case in the actual useful functionality +of views. Note my comment in the http://git-annex.branchable.com/forum/How_to_hide_broken_symlinks/ +forum post; it's not really a good way to hide missing files. +"""]] diff --git a/doc/bugs/annex_view_barfs__fatal__58___Unable_to_add___40__null__41___to_database/comment_4_b7c2abf63c9f350402b3176e028bda03._comment b/doc/bugs/annex_view_barfs__fatal__58___Unable_to_add___40__null__41___to_database/comment_4_b7c2abf63c9f350402b3176e028bda03._comment new file mode 100644 index 0000000000..c991bd3b6a --- /dev/null +++ b/doc/bugs/annex_view_barfs__fatal__58___Unable_to_add___40__null__41___to_database/comment_4_b7c2abf63c9f350402b3176e028bda03._comment @@ -0,0 +1,21 @@ +[[!comment format=mdwn + username="joey" + subject="""comment 4""" + date="2018-05-14T17:30:25Z" + content=""" +It's caused by git submodules with names starting with dots. Not at all +related to the recent null pointer dereference bug in git-annex. + +The .heudiconv/ directory is a submodule and git ls-files lists it; the view +generation code assumes it's a dotfile and asks git to hash it. + + joey@darkstar:~/tmp/1038_hyperface.x>git hash-object -w --stdin-paths --no-filters + .heudiconv + fatal: Unable to add (null) to database + +Views only include dotfiles to make .gitattributes etc work, and they don't +include submodules normally, but they do include dot-dirs too; indeed I see +this repository uses that with a .datalad/ that gets included in the view. +Perhaps .heudiconv/ is a similar dot-dir. Seems that such submodule +dot-dirs ought to be included in the view. +"""]]