Improve docs on 'cp -r', noting the recent 2329 changes

refs ticket:2329

NEWS.rst

@@ -7,9 +7,9 @@ User-Visible Changes in Tahoe-LAFS
 Release 1.10.1 (XXXX-XX-XX)
 '''''''''''''''''''''''''''
 
-Unedited list of all changes after 1.10 and before 0d935e8 06-Jan-2015. This
-list is not yet limited to user-visible ones. It *should* include all tickets
-closed during this time, even minor non-user-visible ones.
+Unedited list of all changes after 1.10.0 and before 0d935e8 06-Jan-2015.
+This list is not yet limited to user-visible ones. It *should* include all
+tickets closed during this time, even minor non-user-visible ones.
 
 - show git branch in version output #1953
 - packaging fixes #1969 #1960
@@ -59,7 +59,7 @@ closed during this time, even minor non-user-visible ones.
 - improve version-number reporting #2340
 - add per-server "(space) Available" column to welcome page #648
 - add public-key auth to SFTP server #1411
-
+- `tahoe cp -r` changes w.r.t. unnamed directories #2329
 
 all tickets noted as closed: 1953 1960 1974 1972 1717 1381 898 1707 1918 1807
 740 1842 1992 2165 1847 2086 2208 2048 2128 2245 1336 2248 2067 712 1800 1966
@@ -72,6 +72,54 @@ all tickets referenced (fixed? not fixed?): 1834 1969 1742 1988 982 1064 1536
 PRs noted as closed: 62 48 57 61 62 63 64 69 73 81 82 84 85 87 91 94 95 96
 103 56 32 50 107 109 114 112 120 122 125 126 133
 
+- "tahoe cp" changes:
+
+There are many "cp"-like tools in the unix world (POSIX /bin/cp, the "scp"
+provided by SSH, rsync). They each behave slightly differently in unusual
+circumstances, generally dealing with copying whole directories at a time,
+into a target which may or may not exist already. The usual question is
+whether the user is referring to the source directory as a whole, or to its
+contents. For example, should "cp -r foodir bardir" create a new directory
+named "bardir/foodir"? Or should it behave more like "cp -r foodir/* bardir"?
+Some tools use the presence of a trailing slash to indicate which behavior
+you want. Others ignore trailing slashes.
+
+"tahoe cp" is no exception to having exceptional cases. This release fixes
+some bad behavior and attempts to establish a consistent rationale for its
+behavior.
+
+The new rule is:
+
+- If the thing being copied is a directory, and it has a name (e.g. it's not
+  a raw tahoe directorycap), then you are referring to the directory itself.
+- If the thing being copied is an unnamed directory (e.g. raw dircap or
+  alias), then you are referring to the contents.
+- Trailing slashes do not affect the behavior of the copy (although putting a
+  trailing slash on a file-like target is an error).
+- The "-r" (--recursive) flag does not affect the behavior of the copy
+  (although omitting -r when the source is a directory is an error).
+- If the target refers to something that does not yet exist:
+  - and if the source is a single file, then create a new file;
+  - otherwise, create a directory.
+
+There are two main cases where the behavior of tahoe-1.10.1 differs from that
+of the 1.10.0 release:
+
+- "cp DIRCAP/file.txt ./local/missing" , where "./local" is a directory but
+  "./local/missing" does not exist. The implication is that you want tahoe to
+  create a new file named "./local/missing" and fill it with the contents of
+  the tahoe-side DIRCAP/file.txt. In 1.10.0, a plain "cp" would do just this,
+  but "cp -r" would do "mkdir ./local/missing" and then create a file named
+  "./local/missing/file.txt". In 1.10.1, both "cp" and "cp -r" create a file
+  named "./local/missing".
+- "cp -r PARENTCAP/dir ./local/missing", where PARENTCAP/dir/ contains
+  "file.txt", and again "./local" is a directory but "./local/missing" does
+  not exist. In both 1.10.0 and 1.10.1, this first does "mkdir
+  ./local/missing". In 1.10.0, it would then copy the contents of the source
+  directory into the new directory, resulting in "./local/missing/file.txt".
+  In 1.10.1, following the new rule of "a named directory source refers to
+  the directory itself", the tool creates "./local/missing/dir/file.txt".
+
 
 
 Release 1.10.0 (2013-05-01)

docs/frontends/CLI.rst

@@ -455,6 +455,21 @@ Command Examples
  all source arguments which are directories will be copied into new
  subdirectories of the target.
 
+ The behavior of ``tahoe cp``, like the regular UNIX ``/bin/cp``, is subtly
+ different depending upon the exact form of the arguments. In particular:
+
+* Trailing slashes indicate directories, but are not required.
+* If the target object does not already exist:
+  * and if the source is a single file, it will be copied into the target;
+  * otherwise, the target will be created as a directory.
+* If there are multiple sources, the target must be a directory.
+* If the target is a pre-existing file, the source must be a single file.
+* If the target is a directory, each source must be a named file, a named
+  directory, or an unnamed directory. It is not possible to copy an unnamed
+  file (e.g. a raw filecap) into a directory, as there is no way to know what
+  the new file should be named.
+
+
 ``tahoe unlink uploaded.txt``
 
 ``tahoe unlink tahoe:uploaded.txt``

src/allmydata/scripts/cli.py

@@ -252,9 +252,14 @@ class CpOptions(FilesystemOptions):
 
     This command still has some limitations: symlinks and special files
     (device nodes, named pipes) are not handled very well. Arguments should
-    probably not have trailing slashes. 'tahoe cp' does not behave as much
-    like /bin/cp as you would wish, especially with respect to trailing
-    slashes.
+    not have trailing slashes (they are ignored for directory arguments, but
+    trigger errors for file arguments). When copying directories, it can be
+    unclear whether you mean to copy the contents of a source directory, or
+    the source directory itself (i.e. whether the output goes under the
+    target directory, or one directory lower). Tahoe's rule is that source
+    directories with names are referring to the directory as a whole, and
+    source directories without names (e.g. a raw dircap) are referring to the
+    contents.
     """
 
 class UnlinkOptions(FilesystemOptions):