Using techniques mentioned here (Pipe output to two different commands) we can split a stdout into multiple processes.
expensive_command | tee >(proc_1) >(proc_2) | proc_3
my problem is this interlaces the output.
Is there a way to copy the stdout but force proc_2 to block until proc_1 finishes?
I'm thinking something like
expensive_command | tee >(proc_1) | wait for EOF | tee >(proc_2) ...
You can use a fifo as a cheap lock. Have proc1 write to it after it completes, and wait until a read from the fifo succeeds before running proc2.
mkfifo cheap_lock
expensive_command | tee >(proc1; echo foo > cheap_lock) \
>(read < cheap_lock; proc2 ) | proc3
(Of course, it's your responsibility to ensure that no other processes try to read from or write to cheap_lock.)
You can create a buffer holder that would release the output once data from input reaches eof like
expensive_command | awk '{ a[i++] = $0 }END{for (i = 0; i in a; ++i) { print a[i] | "tee temp.txt" } }'
Only that awk does not support process substitution.
In bash you can do:
readarray -t lines <(expressive_command | tee >(proc_1))
printf '%s\n' "${lines[#]}" | tee >(proc_2)
Depending on the peak data size of your output from expressive_command or version of your Bash, the command may require adjustments. You can also consider using another language.
Add: You can also use stdbuf. It runs command with modified buffering operations for its standard streams.
Related
my bash for loop looks like:
for i in read_* ; do
cut -f1 $i | sponge $i
sed -i '1 s/^/>/g' $i
sed -i '3 s/^/>ref\n/g' $i
sed -i '4d' $i
sed -i '1h;2H;1,2d;4G' $i
mv $i $i.fasta
done
Are there any methods of speeding up this process, perhaps using GNU parallel?
EDIT: Added input and expected output.
Input:
sampleid 97 stuff 2086 42 213M = 3322 1431
TATTTAGGGAAGATCTGGCCTTCCTACAAGGGAAGGCCAGGGAATTTTCTTCAGAGCAGA
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
TTTTTAGGGAAGATCTGGCCTTCCTACAAGGGAAGGCCAGGGAATTTTCTTCAGAGCAGA
Hopeful output:
>ref
TTTTTAGGGAAGATCTGGCCTTCCTACAAGGGAAGGCCAGGGAATTTTCTTCAGAGCAGA
>sampleid
TATTTAGGGAAGATCTGGCCTTCCTACAAGGGAAGGCCAGGGAATTTTCTTCAGAGCAGA
I used the sed -i '1h;2H;1,2d;4G' $i command to swap lines 2 and 4.
If I read it right, this should create the same result, though it would probably help a LOT if I could see what your input and expected output look like...
awk '{$0=$1}
FNR==1{hd=">"$0; next}
FNR==2{hd=hd"\n"$0;next}
FNR==3{print ">ref\n"$0 > FILENAME".fasta"}
FNR==4{next}
FNR==5{print hd"\n"$0 > FILENAME".fasta"}
' read_*
My input files:
$: cat read_x
foo x
bar x
baz x
last x
curiosity x
$: cat read_y
FOO y
BAR y
BAZ y
LAST y
CURIOSITY y
and the resulting output files:
$: cat read_x.fasta
>ref
baz
>foo
bar
curiosity
$: cat read_y.fasta
>ref
BAZ
>FOO
BAR
CURIOSITY
This runs in one pass with no loop aside from awk's usual internals, and leaves the originals in place so you can check it first. If all is good, all that's left is to remove the originals. For that, I would use extended globbing.
$: shopt -s extglob; rm read_!(*.fasta)
That will clean up the original inputs but not the new outputs.
Same results, three commands, no loops.
I am, or course, making some assumptions about what you are meaning to do that might not be accurate. To get this format in a single sed call -
$: sed -e 's/[[:space:]].*//' -e '1{s/^/>/;h;d}' -e '2{H;s/.*/>ref/}' -e '4x' read_x
>ref
baz
>foo
bar
curiosity
but that's not the same commands you used, so maybe I'm misreading it.
To use this to in-place edit multiple files at a time (instead of calling it in a loop on each file), use -si so that the line numbers apply to each file rather than the stream of records they collectively produce.
DON'T use -is, though you could use -i -s.
$: sed -s -i -e 's/[[:space:]].*//' -e '1{s/^/>/;h;d}' -e '2{H;s/.*/>ref/}' -e '4x' read_*
This still leaves you with the issue of renaming each, but xargs makes that pretty easy in the given example.
printf "%s\n" read_* | xargs -I# mv # #.fasta
addendum
Using the file you gave in the OP, assuming every file is the same general structure and exactly 4 lines -
$: cat file_0 # I made files 0 through 7, but with same data
sampleid 97 stuff 2086 42 213M = 3322 1431
TATTTAGGGAAGATCTGGCCTTCCTACAAGGGAAGGCCAGGGAATTTTCTTCAGAGCAGA
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
TTTTTAGGGAAGATCTGGCCTTCCTACAAGGGAAGGCCAGGGAATTTTCTTCAGAGCAGA
$: sed -Esi '1{s/^([^[:space:]]+).*/>\1/;h;s/.*/>ref/}; 3x;' file_?
$: cat file_0 # used a diff on each, worked on all at once
>ref
TATTTAGGGAAGATCTGGCCTTCCTACAAGGGAAGGCCAGGGAATTTTCTTCAGAGCAGA
>sampleid
TTTTTAGGGAAGATCTGGCCTTCCTACAAGGGAAGGCCAGGGAATTTTCTTCAGAGCAGA
Breakout:
-Esi Extended pattern matching, separate file linecounts, in-place edits
1{...}; Collectively do these commands, in order, only on every line 1
s/^([^[:space:]]+).*/>\1/ add leading > but strip everything after any whitespace
h store the resulting >\1 line in the hold buffer
s/.*/>ref/ then replace the whole line with a literal >ref
`3x' swap line 3 with the value in the hold buffer from line 1
file_? I used a glob to supply the appropriate list of files all at once.
Doing same with awk:
$: awk 'FNR==1{id=">"$1; print ">ref" >FILENAME".fasta"; next} FNR==3{print id > FILENAME".fasta"; next} {print $0 > FILENAME".fasta"}' file_?
Then you can do file management as above with the xargs/mv for the sed or the shopt/rm for the awk - or we could add a little organizational work in awk if you like. Consider this:
awk 'BEGIN { system(" mkdir -p done ") }
FNR==1 { id=">"$1; print ">ref" > FILENAME".fasta"; next } # skip printing original
FNR==3 { print id > FILENAME".fasta"; next } # skip printing original
{ print $0 > FILENAME".fasta" } # every line NOT skipped
FNR==4 { close(FILENAME); close(FILENAME".fasta");
system("mv " FILENAME " done/")
}' file_?
Then if there are any problems, it's easy to delete the fasta's, move the originals back, adjust the code, and try again. If everything is ok, it's fast and easy to rm -fr done, yes?
Note that I really only added the mkdir inside a system call in the awk to show that you can, and to keep from having to manually do it separately if you have to run a few iterations or move it all into a wrapper script, etc.
The code in the question runs multiple subprocesses (cut, sponge, sed four times, and mv) for each file that is processed. Running subprocesses is relatively slow, so you can speed up the code significantly by reducing the number of them.
This Shellcheck-clean code is one way to do it:
#! /bin/bash -p
old_files=()
for f in read_* ; do
readarray -t lines <"$f"
printf '>ref\n%s\n>%s\n%s\n' \
"${lines[3]}" "${lines[0]%%[[:space:]]*}" "${lines[1]}" >"$f.fasta"
old_files+=( "$f" )
done
rm -- "${old_files[#]}"
This runs no subprocesses when processing individual files. It just reads the lines of the old file into an array using the built-in readarray command and writes to the new file using the built-in printf.
See Removing part of a string (BashFAQ/100 (How do I do string manipulation in bash?)) for an explanation of the %% in ${lines[0]%%[[:space:]]*}.
To avoid running rm for each file, the code keeps a list of files to be deleted and removes all of them at the end. If you try the code, consider commenting the rm line until you are very confident that the rest of the code is doing what you want.
Suppose I have the following command in bash:
one | two
one runs for a long time producing a stream of output and two performs a quick operation on each line of that stream, but two doesn't work at all unless the first value it reads tells it how many values to read per line. one does not output that value, but I know what it is in advance (let's say it's 15). I want to send a 15\n through the pipe before the output of one. I do not want to modify one or two.
My first thought was to do:
echo "$(echo 15; one)" | two
That gives me the correct output, but it doesn't stream through the pipe at all until the command one finishes. I want the output to start streaming right away through the pipe, since it takes a long time to execute (months).
I also tried:
echo 15; one | two
Which, of course, outputs 15, but doesn't send it through the pipe to two.
Is there a way in bash to pass '15\n' through the pipe and then start streaming the output of one through the same pipe?
You just need the shell grouping construct:
{ echo 15; one; } | two
The spaces around the braces and the trailing semicolon are required.
To test:
one() { sleep 5; echo done; }
two() { while read line; do date "+%T - $line"; done; }
{ printf "%s\n" 1 2 3; one; } | two
16:29:53 - 1
16:29:53 - 2
16:29:53 - 3
16:29:58 - done
Use command grouping:
{ echo 15; one; } | two
Done!
You could do this with sed:
Example 'one' script, emits one line per second to show it's line buffered and running.
#!/bin/bash
while [ 1 ]; do
echo "TICK $(date)"
sleep 1
done
Then pipe that through this sed command, note that for your specific example 'ArbitraryText' will be the number of fields. I used ArbitraryText so that it's obvious that this is the inserted text. On OSX, -l is unbuffered with GNU Sed I believe it's -u
$ ./one | sed -l '1i\
> ArbitraryText
> '
What this does is it instructs sed to insert one line before processing the rest of your file, everything else will pass through untouched.
The end result is processed line-by-line without chunk buffering (or, waiting for the input script to finish)
ArbitraryText
TICK Fri Jun 28 13:26:56 PDT 2013
...etc
You should be able to then pipe that into 'two' as you would normally.
How to I concatenate stdin to a string, like this?
echo "input" | COMMAND "string"
and get
inputstring
A bit hacky, but this might be the shortest way to do what you asked in the question (use a pipe to accept stdout from echo "input" as stdin to another process / command:
echo "input" | awk '{print $1"string"}'
Output:
inputstring
What task are you exactly trying to accomplish? More context can get you more direction on a better solution.
Update - responding to comment:
#NoamRoss
The more idiomatic way of doing what you want is then:
echo 'http://dx.doi.org/'"$(pbpaste)"
The $(...) syntax is called command substitution. In short, it executes the commands enclosed in a new subshell, and substitutes the its stdout output to where the $(...) was invoked in the parent shell. So you would get, in effect:
echo 'http://dx.doi.org/'"rsif.2012.0125"
use cat - to read from stdin, and put it in $() to throw away the trailing newline
echo input | COMMAND "$(cat -)string"
However why don't you drop the pipe and grab the output of the left side in a command substitution:
COMMAND "$(echo input)string"
I'm often using pipes, so this tends to be an easy way to prefix and suffix stdin:
echo -n "my standard in" | cat <(echo -n "prefix... ") - <(echo " ...suffix")
prefix... my standard in ...suffix
There are some ways of accomplish this, i personally think the best is:
echo input | while read line; do echo $line string; done
Another can be by substituting "$" (end of line character) with "string" in a sed command:
echo input | sed "s/$/ string/g"
Why i prefer the former? Because it concatenates a string to stdin instantly, for example with the following command:
(echo input_one ;sleep 5; echo input_two ) | while read line; do echo $line string; done
you get immediatly the first output:
input_one string
and then after 5 seconds you get the other echo:
input_two string
On the other hand using "sed" first it performs all the content of the parenthesis and then it gives it to "sed", so the command
(echo input_one ;sleep 5; echo input_two ) | sed "s/$/ string/g"
will output both the lines
input_one string
input_two string
after 5 seconds.
This can be very useful in cases you are performing calls to functions which takes a long time to complete and want to be continuously updated about the output of the function.
You can do it with sed:
seq 5 | sed '$a\6'
seq 5 | sed '$ s/.*/\0 6/'
In your example:
echo input | sed 's/.*/\0string/'
I know this is a few years late, but you can accomplish this with the xargs -J option:
echo "input" | xargs -J "%" echo "%" "string"
And since it is xargs, you can do this on multiple lines of a file at once. If the file 'names' has three lines, like:
Adam
Bob
Charlie
You could do:
cat names | xargs -n 1 -J "%" echo "I like" "%" "because he is nice"
Also works:
seq -w 0 100 | xargs -I {} echo "string "{}
Will generate strings like:
string 000
string 001
string 002
string 003
string 004
...
The command you posted would take the string "input" use it as COMMAND's stdin stream, which would not produce the results you are looking for unless COMMAND first printed out the contents of its stdin and then printed out its command line arguments.
It seems like what you want to do is more close to command substitution.
http://www.gnu.org/software/bash/manual/html_node/Command-Substitution.html#Command-Substitution
With command substitution you can have a commandline like this:
echo input `COMMAND "string"`
This will first evaluate COMMAND with "string" as input, and then expand the results of that commands execution onto a line, replacing what's between the ‘`’ characters.
cat will be my choice: ls | cat - <(echo new line)
With perl
echo "input" | perl -ne 'print "prefix $_"'
Output:
prefix input
A solution using sd (basically a modern sed; much easier to use IMO):
# replace '$' (end of string marker) with 'Ipsum'
# the `e` flag disables multi-line matching (treats all lines as one)
$ echo "Lorem" | sd --flags e '$' 'Ipsum'
Lorem
Ipsum#no new line here
You might observe that Ipsum appears on a new line, and the output is missing a \n. The reason is echo's output ends in a \n, and you didn't tell sd to add a new \n. sd is technically correct because it's doing exactly what you are asking it to do and nothing else.
However this may not be what you want, so instead you can do this:
# replace '\n$' (new line, immediately followed by end of string) by 'Ipsum\n'
# don't forget to re-add the `\n` that you removed (if you want it)
$ echo "Lorem" | sd --flags e '\n$' 'Ipsum\n'
LoremIpsum
If you have a multi-line string, but you want to append to the end of each individual line:
$ ls
foo bar baz
$ ls | sd '\n' '/file\n'
bar/file
baz/file
foo/file
I want to prepend my sql script with "set" statement before running it.
So I echo the "set" instruction, then pipe it to cat. Command cat takes two parameters : STDIN marked as "-" and my sql file, cat joins both of them to one output. Next I pass the result to mysql command to run it as a script.
echo "set #ZERO_PRODUCTS_DISPLAY='$ZERO_PRODUCTS_DISPLAY';" | cat - sql/test_parameter.sql | mysql
p.s. mysql login and password stored in .my.cnf file
Is there an idiomatic analog to Ruby's Object#tap for Unix command pipelines?
Use case: within a pipeline I want to execute a command for its side effects but return the input implicitly so as to not break the chaining of the pipeline. For example:
echo { 1, 2, 3 } |
tr ' ' '\n' |
sort |
tap 'xargs echo' | # arbitrary code, but implicitly return the input
uniq
I'm coming from Ruby, where I would do this:
[ 1, 2, 3 ].
sort.
tap { |x| puts x }.
uniq
The tee command is similar; it writes its input to standard output as well as one or more files. If that file is a process substitution, you get the same effect, I believe.
echo 1 2 3 | tr ' ' '\n' | sort | tee >( **code** ) | uniq
The code in the process substitution would read from its standard input, which should be the same thing that the call to uniq ends up seeing.
So my problem is that I need to have the output of running the command dumped to the screen and also capture it in a variable in a ruby script. I know that I can do the second part like this:
some_variable = `./some_kickbutt`
But my problem is that I need it to still print to the console as Hudson captures that output and records it for posterity's sake.
thanks in advance for any ideas...
Just tee the stdout stream to stderr like so:
ruby -e 'var = `ls | tee /dev/stderr`; puts "\nFROM RUBY\n\n"; puts var' | nl
ruby -e 'var = `ls | tee /dev/stderr`; puts "\nFROM RUBY\n\n"; puts var' 2>&1 | nl