The important thing to note here is the tag on the question. This behaviour is specific to GNOME Terminal and any other terminal emulators that are built upon libvte. You won't see this in Xterm, or in Unicode RXVT, or in the terminal emulator built into the Linux kernel, or on the FreeBSD console.
What happens in general is this.
clear command looks at terminfo/termcap and issues appropriate control sequences.
- If the terminfo/termcap entry has an
E3 capability it, it first writes out that. This issues control sequences to clear the scrollback buffer. This and the history behind it are documented in detail in the Dickey ncurses manual page for the
- It then uses the
clear capability to clear the visible screen.
- The control sequences in the terminfo/termcap entry are determined by the terminal type; but, with the exceptions of the (nowadays rare) terminals that use FormFeed to clear the screen (which DEC VTs and their imitators do not), they are either just plain old ECMA-48 control sequences or extensions thereto. For examples:
- The terminal emulator acts upon the control sequences. As defined by ECMA-48 and the Xterm extension to it:
H (CUP) homes the cursor.
J (ED 0) or just CSI
J erases from the current cursor position to the end of the screen.
J (ED 2) erases the whole screen.
J (ED 3) erases the scrollback buffer.
When it comes to GNOME Terminal in particular:
- The terminal type is properly
gnome, but some people leave it erroneously set to
gnome terminfo entry does not define an
E3 capability, and on many systems — still! — neither does the
xterm entry as this has not percolated down from Dickey terminfo. So
clear just writes out the contents of the
- The contents of the
clear capability for those terminfo entries are the control sequences to home the cursor followed by erase the whole screen.
- But GNOME Terminal does not implement erase the whole screen correctly. More specifically, the library that it is based upon, libvte, does not do that in the code of its
VteTerminalPrivate::seq_clear_screen() function. Rather, libvte scrolls the screen down an entire screen's worth of blank lines, and moves the cursor position to the first of those blank lines.
This is why you see what you see. libvte is not erasing the whole screen when told to. Rather it is doing something that has a superficial resemblance to that, until one does exactly what the questioner has done here: scroll the terminal window back to look at the scroll back buffer. Then the difference is blatant.
On other terminal emulators such as Xterm and Unicode RXVT, the ED 2 control sequence really does erase the screen, erasing every position on the screen in place, from the top down, and not altering the scrollback buffer. But on libvte terminal emulators, it just pushes the current screen up into the scrollback buffer and adds a screen's worth of blank lines. The prior screen contents are not erased but shifted into the scrollback buffer.
And if you run the
clear command twice, it adds two screen's worth of blank lines. If your scroll back buffer is large enough, you can still find the original screen contents, simply further up in the scrollback buffer.