cp -r * except dont copy any .pdf files - copy a directory subtree while excluding files with a given extension - bash

Question

Editor's note: In the original form of the question the aspect of copying an entire subtree was not readily obvious.
How do I copy all the files from one directory subtree to another but omit all files of one type?
Does bash handle regex?
Something like: cp -r !*.pdf /var/www/ .?
EDIT 1
I have a find expression: find /var/www/ -not -iname "*.pdf"
This lists all the files that I want to copy. How do I pipe this to a copy command?
EDIT 2
This works so long as the argument list is not too long:
sudo cp `find /var/www/ -not -iname "*.pdf"` .
EDIT 3
One issue though is that I am running into issues with losing the directory structure.

Answer 1

Bash can't help here, unfortunately.
Many people use either tar or rsync for this type of task because each of them is capable of recursively copying files, and each provides an --exclude argument for excluding certain filename patterns. tar is more likely to be installed on a given machine, so I'll show you that.
Assuming you are currently in the destination directory, the shell command:
tar -cC /var/www . | tar -x
will copy all files from /var/www into the current directory recursively.
To filter out the PDF files, use:
tar -cC /var/www --exclude '*.pdf' . | tar -x
Multiple --exclude arguments can be given, so:
tar -cC /var/www --exclude '*.pdf' --exclude '*.txt' . | tar -x
would exclude .txt files as well.

Answer 2

K. A. Buhr's helpful answer is a concise solution that reflects the intent well and is easily extensible if multiple extensions should be excluded.
Trying to do it with POSIX utilities and POSIX-compliant options alone requires a slightly different approach:
cp -pR /var/www/. . && find . -name '*.pdf' -exec rm {} +
In other words: copy the whole subtree first, then remove all *.pdf files from the destination subtree.
Note:
-p preserves the original files' attributes in terms of file timestamps, ownership, and permission bits (tar appears to do that by default); without -p, the copies will be owned by the current user and receive new timestamps (though the permission bits are preserved).
Using cp has one advantage over tar: you get more control over how symlinks among the source files are handled, via the -H, -L, and -P options - see the POSIX spec. for cp.
tar invariably seems to copy symlinks as-is.
-R supersedes the legacy -r option for cp, as the latter's behavior with non-regular files is ill-defined - see the RATIONALE section in the POSIX spec. for cp
Neither -iname for case-insensitive matching nor -delete are part of the POSIX spec. for find, but both GNU find and BSD/macOS find support them.
Note how source path /var/www/. ends in /. to ensure that its contents are copied to the destination path (as opposed to putting everything into a www subfolder).
With BSD cp, /var/www/ (trailing /) would work too, but GNU cp treats /var/www and /var/www/ the same.
As for your questions and solution attempts:
Does bash handle regex?
In the context of filename expansion (globbing), Bash only understands patterns, not regexes (Bash does have the =~ regex-matching operator for string matching inside [[ ... ]] conditionals, however).
As a nonstandard extension, Bash implements the extglob shell option, which adds additional constructs to the pattern-matching notation to allow for more sophisticated matching, such as !(...) for negating matchings, which is what you're looking for.
If you combine that with another nonstandard shell option, globstar (**, Bash v4+), you can construct a single pattern that matches all items except a given sub-pattern across an entire subtree:
/var/www/**/!(*.pdf)
does find all non-PDF filesystem items in the subtree of /var/www/.
However, combining that pattern with cp won't work as intended: with -R, any subdirs. are still copied in full; without -R, subdirs. are ignored altogether.
Caveats:
By default, patterns (globs) ignore hidden items unless explicitly matched (* will only match non-hidden items). To include them, set shell option dotglob first.
Matching is case-sensitive by default; turn on shell option nocaseglob to make it case-insensitive.
find /var/www/ -not -iname "*.pdf" in essence yields the same as the extended glob above, except with case-insensitive matching, hidden items invariably included, and the output paths (generally) not in the same order.
However, copying the output paths to their intended destination is the nontrivial part: you'd have to construct analogous subdirs. in the destination dir. on the fly, and you'd have to do so for each input path separately, which will also be quite slow.
Your own attempt, sudo cp `find /var/www/ -not -iname "*.pdf"` ., falls short in several respects:
As you've discovered yourself, this copies all matching items into a single destination directory.
The output of the command substitution, `...`, is subject to shell expansions, namely word-splitting and filename expansion, which may break the command, notably with filenames with embedded spaces.
Note: As written, all destination items will be owned by the root user.

Answer 3

Edit As per #mklement0's comment below, these solutions are not suitable for directory tree recursion--they will only work on one directory, as per the OP's original form of the question.
#rorschach. Yes, you can do this.
Using cp:
Set your Bash shell's extglob option and type:
shopt -s extglob #You can set this in your shell startup to enable it by default
cp /var/www/!(*.pdf) .
If you wish to turn off (unset) this (or any other) shell option, use:
shopt -u extglob #or whatever shell option you wish to unset
Using find
If you prefer using find, you can use xargs to execute the operation you would like Bash to perform:
find /var/www/ ! -iname "*.pdf" -maxdepth 1 | xargs -I{} cp {} .