-
-
Save earthgecko/3089509 to your computer and use it in GitHub Desktop.
#!/bin/bash | |
# bash generate random alphanumeric string | |
# | |
# bash generate random 32 character alphanumeric string (upper and lowercase) and | |
NEW_UUID=$(cat /dev/urandom | tr -dc 'a-zA-Z0-9' | fold -w 32 | head -n 1) | |
# bash generate random 32 character alphanumeric string (lowercase only) | |
cat /dev/urandom | tr -dc 'a-z0-9' | fold -w 32 | head -n 1 | |
# Random numbers in a range, more randomly distributed than $RANDOM which is not | |
# very random in terms of distribution of numbers. | |
# bash generate random number between 0 and 9 | |
cat /dev/urandom | tr -dc '0-9' | fold -w 256 | head -n 1 | head --bytes 1 | |
# bash generate random number between 0 and 99 | |
NUMBER=$(cat /dev/urandom | tr -dc '0-9' | fold -w 256 | head -n 1 | sed -e 's/^0*//' | head --bytes 2) | |
if [ "$NUMBER" == "" ]; then | |
NUMBER=0 | |
fi | |
# bash generate random number between 0 and 999 | |
NUMBER=$(cat /dev/urandom | tr -dc '0-9' | fold -w 256 | head -n 1 | sed -e 's/^0*//' | head --bytes 3) | |
if [ "$NUMBER" == "" ]; then | |
NUMBER=0 | |
fi |
Also a much more sketchy but works way
head -c 1024 /dev/urandom | base64 | tr -cd "[:upper:][:digit:]" | head -c 32
tested on macOS 10.15.7
it does not promise a success, but after generating 1024 chars on random, the chance of it not containing 32 upper case and number is astronomically slim.
If you are curious, the chance of not generating 512 char and only have <32 upper&digit is 2.051*10^-195. and the chance of failing of 1024 is ....... ok, mathmetica literately tells me it "is too small to represent as a normalized machine number" 😆 and it is about 10^-400
and btw it takes 7ms to execute; on my machine 🤣
Which one of all the different options given above is the fastest?
As an example - if one wants to generate 10000 lines each 16 random chars long using 0-9a-zA-Z as a dictionary, which method would be the fastest?
If you need add spéciale caractère
echo NEW_UUID=$(cat /dev/urandom | tr -dc 'a-zA-Z0-9&-@' | fold -w 10 | head -n 1)
Generate UUID-like strings:
$ od -x /dev/urandom | head -1 | awk '{OFS="-"; print $2$3,$4,$5,$6,$7$8$9}'
# 14e9710d-e38a-fbba-4a97-a8972fbf2fd0
This bypasses any neccessity to installsoftware like uuidgen or external modules for Perl or Python. Tested on macOs and Linux.
this is great! thanks!
cat /dev/urandom | tr -dc 'a-zA-Z0-9' | fold -w 32 | head -n 1
(2nd command) is not generating lowercase only while this one yes
cat /dev/urandom | tr -dc 'a-z0-9' | fold -w 32 | head -n 1
alternative on systems that don't have many packages (openwrt old versions)
dd if=/dev/urandom bs=10 count=20 | tr -dc 'a-zA-Z0-9' | xargs echo
Adjust bs
and count
as you wish.
The following seems to work fine both on Linux and macOS (4 alphanum characters):
cat /dev/urandom | env LC_CTYPE=C tr -dc 'a-z0-9' | fold -w 4 | head -n 1
Thanks
Can someone help me to generate random letters only?
@UdaniWijesinghe just remove the 0-9
# bash generate random 32 character alphabetic only string (upper and lowercase) and
NEW_UUID=$(cat /dev/urandom | tr -dc 'a-zA-Z' | fold -w 32 | head -n 1)
# bash generate random 32 character alphabetic string (lowercase only)
cat /dev/urandom | tr -dc 'a-z' | fold -w 32 | head -n 1
on macOS (/dev/urandom does not spits out any valid CTYPE, is this M1 related?)
# generate single letter in range a-z
head /dev/urandom | base64 | tr -dc 'a-z' | fold -w 1 | head -n 1
The base64
approach is interesting to make every byte count (for a-zA-Z and 0-9). only I am not really sure if base64 maps the bytes uniformingly on a-zA-Z0-9 . I didn't check so far, has anyone experience?
@pier4r that's a fair question, though base64 breaks streams in 6bits chars, given that the source of randomnes should minimize the correlation between the previous N bits and the current ones, the only reduction of information is due to the tr step, which is intrinsic in the goal of the opertions (generate range of alphachars/numbers/both)
base64
should be the best thing you can do. This way no incoming bit will be ignored. tr
on the other hand is not the best thing in my opinion. But it also would be a bit more complicated if you always want to make every bit count.
When generating one or two 10-20 char length strings which method you pick up probably is not that important.
How about putting the question differently - which method would be least resource-intensive and won't require installation of additional tool? Let's imagine you need to generate 5000 strings each 160 char length.
invisible see https://gist.github.com/earthgecko/3089509#gistcomment-3827754 (achieved on a very small footprint OS, openwrt)
@pier4r Thanks. But what performance would look like compared to other methods?
@invisible999 don't know, one needs to make a benchmark with the various proposed methods (and likely others published elsewhere)
how do i generate the string cat /dev/urandom | tr -dc 'a-zA-Z0-9' | fold -w 32 | head -n 1 but have the first half as dedicated set characters
Exactly what I need ;)