<stránka

	xmlns="https://trac.frantovo.cz/xml-web-generator/wiki/xmlns/strana"

	xmlns:m="https://trac.frantovo.cz/xml-web-generator/wiki/xmlns/makro">

	<nadpis>Streamlets preview</nadpis>

	<perex>an early example of streamlets in relpipe-in-filesystem</perex>

	<text xmlns="http://www.w3.org/1999/xhtml">

		<p>

			<em>This is an early preview published at 2020-01-17 before the v0.15 release.</em>

		</p>

		<p>

			First prepare some files:

		</p>

		<m:pre jazyk="shell"><![CDATA[$ wget --xattr https://upload.wikimedia.org/wikipedia/commons/d/d4/HURD_Live_CD.png

$ wget --xattr https://sane-software.globalcode.info/v_0/ssm.en.pdf

$ wget --xattr https://alt2xml.globalcode.info/sql-api_alt2xml_talk_2014.pdf

$ ls -1

HURD_Live_CD.png

search.sh

sql-api_alt2xml_talk_2014.pdf

ssm.en.pdf]]></m:pre>

		<p>

			Collect metadata (file path, extended attributes, image size, number of PDF pages, number of text lines, OCR recognized text extracted from images and plain-text extracted from PDF files),

			filter the results (do restriction), select only certain attributes (do projection)

			and format result as a table:

		</p>

		<m:pre jazyk="shell"><![CDATA[find -print0 \

	| relpipe-in-filesystem \

		--file path \

		--xattr xdg.origin.url --as 'url' \

		--streamlet exiftool \

			--option 'attribute' 'PNG:ImageWidth'  --as 'width' \

			--option 'attribute' 'PNG:ImageHeight' --as 'height' \

			--option 'attribute' 'PDF:PageCount'   --as 'page_count' \

		--streamlet lines_count \

		--streamlet tesseract \

			--option 'language' 'eng' \

			--as 'ocr_text' \

		--streamlet pdftotext --as 'pdf_text' \

	| relpipe-tr-awk \

		--relation filesystem \

		--where 'path ~ /\.sh$/ || url ~ /alt2xml\.globalcode\.info/ || ocr_text ~ /GNU/ || pdf_text ~ /Sane/' \

	| relpipe-tr-cut filesystem 'path|url|width|height|page_count|lines_count' \

	| relpipe-out-tabular

# if too wide, add: | less -RSi]]></m:pre>

		<p>

			Which will print:

		</p>

		<m:pre jazyk="text"><![CDATA[filesystem:

 ╭─────────────────────────────────┬──────────────────────────────────────────────────────────────────────┬────────────────┬─────────────────┬─────────────────────┬───────────────────────╮

 │ path                   (string) │ url                                                         (string) │ width (string) │ height (string) │ page_count (string) │ lines_count (integer) │

 ├─────────────────────────────────┼──────────────────────────────────────────────────────────────────────┼────────────────┼─────────────────┼─────────────────────┼───────────────────────┤

 │ ./HURD_Live_CD.png              │ https://upload.wikimedia.org/wikipedia/commons/d/d4/HURD_Live_CD.png │ 720            │ 400             │                     │                     8 │

 │ ./ssm.en.pdf                    │ https://sane-software.globalcode.info/v_0/ssm.en.pdf                 │                │                 │ 6                   │                   568 │

 │ ./sql-api_alt2xml_talk_2014.pdf │ https://alt2xml.globalcode.info/sql-api_alt2xml_talk_2014.pdf        │                │                 │ 21                  │                   696 │

 │ ./search.sh                     │                                                                      │                │                 │                     │                    21 │

 ╰─────────────────────────────────┴──────────────────────────────────────────────────────────────────────┴────────────────┴─────────────────┴─────────────────────┴───────────────────────╯

Record count: 4]]></m:pre>

		<p>

			How it looks in the terminal:

		</p>

		<m:img src="img/streamlets-preview.png"/>

		<p>

			OCR and PDF text extractions (and also other metadata extractions) are done on-the-fly in the pipeline.

			Especially the OCR may take some time, so it is usually better in such case to break the pipe in the middle, 

			redirect intermediate result to a file (serves like an index or cache) and then use it multiple times 

			(just <code>cat</code> the file and continue the original pipeline; BTW: multiple files can be simply concatenated, the format is designed for such use). 

			But in most cases, it is not necessary and we work with live data.

		</p>

		<p>

			Please note that this is really fresh, it has not been released and can be seen only in the Mercurial repository.

			The streamlets used can be seen here: <a href="https://hg.globalcode.info/relpipe/relpipe-in-filesystem.cpp/file/tip/streamlet-examples">streamlet-examples</a>. 

			And even the upcoming release v0.15 is still a development version (it will work, but the API might change in future – until we release v1.0 which will be stable and production ready).

		</p>

		<p>

			Regarding performance:

			currently it is parallelized only over attributes (each streamlet instance runs in a separate process). 

			In v0.15 it will be parallelized also over records (files in this case).

		</p>

	</text>

</stránka>