It's a pity attitudes to climate change research per se are so polarised. There was a paper in a climate modelling conference where the US author was only half joking when he used an amusing title on his paper.
That might be a very dangerous proposal indeed, not least because people might think it would work and actually test it without having any means to reverse it should it be harmful, but also because as per my previous post global warming might cause severe global cooling. (archive-mode proof link here)
What can global warming do? According to the widely accepted Rahmstorf bifurcation model of Atlantic thermohaline circulation, small increases above a threshold level in freshwater precipitation over the North Atlantic region can trigger a long-term Gulf Stream shutdown, a.k.a. an Ice Age.
A super-threshold increase in freshwater supply to the North Atlantic area is one of the most stable effects predicted in simulations by coupled ocean-atmosphere models of climate for any future increased levels of atmospheric CO2.
Would an ice age be so bad? Well, which of the two states below do you prefer?
What does an ice age do? On land, agricultural productivity in the major food producers like the US and N.Europe would collapse due to freezing temperatures. Reduced ocean circulation would also cut precipitation in other regions of the world, creating droughts. In the sea, the much reduced ocean circulation would be unable to replenish marine nutrients in vital high-latitude oceans, i.e., marine productivity and fishing harvests would also collapse.
You don't say which country you're asking about but in the UK and several other European countries too, a land owner is legally obligated through national laws to provide access to any utility company that requests land use and the land owner is entitled in return to a "reasonable" fee, which is usually negotiable based on a notional market rate. The UK legal term for such access rights is a "wayleave".
You might also be interested in learning written Chinese using a really excellent Chinese character tutor called "Hanzim" which is free GPLd software -- ~I posted more information about Hanzim here.
Hanzim, the Chinese character tutor, is also very good for learning the Japanese Kanji characters and multi-character combinations, which are based on traditional Chinese characters. Hanzim doesn't help with learning the Katakana or Hiragana, which are alphabetic not ideographic, but the alphabetic characters are simple to learn. There is also no Japanese pronunciation provided although it'd be easy to add to Hanzim's database.
Adrian Robert of UCSD has created Hanzim, a truly excellent tutor for people learning to read written Chinese. Hanzim is GPLd and has a nice GUI in Tcl/Tk 8.1.
I think the Hanzim project is an outstanding achievement. I wouldn't be surprised if the author starts getting a lot of press recognition for his work, and also one of the Free Software Awards for educational software. Afterall, he has created, with some help, a dictionary with over 6000 individual Chinese characters and over 18000 Chinese character combinations, each with English translations as well as a cross-referenced Chinese radical lookup facility.
In the UK you can opt out of paper junk mail by registering your name and address with the Mailing Preference Service. After I registered I got no more paper junk mail addressed to me. Occasionally I get junk mail sent to my address which have no name on them.
Mailing Preference Service
Freepost 22 London W1E 7EZ
I was wondering whether there is a similar service in other countries?
Have they included the massive extra cost of building an earthquake-proof tunnels? The whole region between Alaska and the Siberian Peninsula is located over major fault planes. Geologist's have concluded there was a massive quake a few centuries ago that was much bigger than 8.5 and possibly as big as 9.7 (more than one order of magnitude more energy than an 8.5)
Nobody really knows how big the quakes can get out there.
The JPEG2000 copy control requirements were apparently added only recently as part of the application requirements but the wider goal is to "ensure that appropriate tools and technologies are adopted in JPEG2000 standard to fullfil [the requirements]". They are not mentioned at all in the full draft ISO/IEC specification Final CD published on 16th March 2000.
"anyone implementing software according to the description available in this FCD, risks not being compliant with the final JPEG2000 International Standard (IS), which is due to be published some time in 2001 as IS15444-1.
So don't expect any projects either to implement JPEG2000 yet or to be able to get JPEG2000 patent licenses (see below) to do so yet.
2. Another important issue (stated here) is that the royalty-free fee-free JPEG2000 patent licenses may apply only to conforming implementations. One such requirement of conforming implementations is to have copy control implemented including methods
1) To protect access to the image
2) To
identify the image, source or owner in a secure way that cannot be removed by unauthorised parties.
3) To indicate integrity (images that are not allowed to be edited).
It is not clear how the patent holders will interpret these requirements for open-source implementations that want to use a GPL license (as for Gimp) which requires the whole software to be modifiablefor any purpose, potentially in ways which could violate the conformance requirements.
I started this thread talking about anti-aliasing of color text
images in the context of using uncalibrated CRT monitors. Keith's point,
and mine too, was that for anti-aliasing of color text images to be correct,
color interpolation must be computed in a linear intensity space. At this point, the topic was still anti-aliasing of color text images.
It was therefore quite a surprise when you replied to Keith Packard's followup
comment saying,
"[it would be wrong]
to change the DAC converters [aren't DACs always converters?] because adjusting the monitor to be
'linear' would spread the samples at the dark end out so much that the steps
between them would be easily visible, while compressing the samples at the
white end so close together that thye cannot be distinguished"
You totally misunderstood Keith's point, and went along a tangent talking about
brightness steps which is not obviously relevant to anti-aliasing of color
text images. Let's stay on my topic of anti-aliasing in a linearised
intensity space.
As you likely know there are two obvious ways of getting an approximately
linear intensity space for a typical CRT connected to a graphics card with
programmable DACs:
1. Adjust the DACs to get a linearised intensity response
by programming them with inverse gamma lookup tables.
2. Adjust the gray
levels in each color channel of an image by mapping gray-level through inverse
gamma lookup tables.
Method 1 is fast but has the disadvantage, acknowledged in Keith's comment,
of not producing a linearised intensity space for images which already have
gamma-corrected gray-levels. Method 2 is slow but can be applied only where
necessary, i.e. only to gamma-uncorrected images.
Regardless of which method is used, the result is a linearised intensity
space. Doing anti-aliasing in any other intensity space such as the exponential
intensity space typical of an uncalibrated CRT would be wrong.
-- -- -- -- --
Oops, I meant the Chevreul illusion, not Mach bands! The Chevreul
illusion is the apparent increase in brightness contrast between adjacent steps
in a "staircase" image containing steps of uniform gray-level which increases from each step to the next step. The reason why any steps
are visible in gray-level staircases with more than about 30 steps is the Chevreul effect. If large black gaps were inserted between adjacent steps
in a staircase image you would not be able to see any brightness difference
between adjacent steps. That's because the gaps between the patches prevent
border contrast effects such as the Chevreul effect which are necessary for us to estimate relative
intensity accurately -- do an experiment and see for yourself.
Mach bands: As an aside, your explanation is a bit over-simplified.
Mach bands are generally visible for a stimulus composed of a blurred step edge
between adjacent white and black areas provided the edge is neither too smooth
nor too sharp, but there's no need for discontinuous first or higher-order
derivatives of intensity -- do an experiment and see for yourself.
Linearity of digital cameras:
"the circuitly in all digital cameras forces that linear output
through a lookup table to convert it to a nonlinear value"
That's not
true. I've done spectrophotometric calibrations of many devices including
famous-brand CCD-based camera and scanner systems; none of them has
had any significant non-linearity in the mapping of intensity to gray-level. There are rare, special-purpose CMOS-based sensors with deliberately non-linear
analog responses like Mead's artificial retina but they are not used in any
consumer camera systems. The old "plumbicon" cameras based on vacuum tubes
were the last common imaging devices to suffer a severe non-linearity --
an exponential response similar to CRTs due to their use of phosphors.
You're right you need to apply a logarithmic mapping to gray level
g'=log(g) for a CRT display but your explanation is wrong. The real
reason is because a display should have a linear output light intensity
with respect to both input voltage and gray-level.
A logarithmic mapping
linearises the intensity space of a CRT but it does not equalise the
contrast differences in a gray-level staircase, which actually requires an
exponential intensity space. To see why you'd need an exponential space for contrast equalisation, consider the gray-level staircase image with y as absolute intensity,
dy/y as relative intensity between adjacent steps (aka brightness contrast) and x as
the gray-level. To get the relative intensity to be a constant we have
dy/dx=y.C implies x=log_e(y)/C = log_K(y)
where C=log_e(K), so y=K^x. In plain English this means a staircase image is contrast-equalised only if absolute intensity
is an exponential function of gray-level. Of course a CRT without
gamma correction already has absolute intensity as an exponential function
of gray-level, so gray-levels don't need to be changed by any gray-level
mapping. A logarithmic gray-level mapping would make brightness contrasts decrease across the staircase.
Your dynamic range is too small. The eye actually has a massive 10^12:1
global dynamic range. The pupil doesn't account for much dynamic range;
the range of pupil diameter is 2-7mm giving only a 38:1 range of light
attenuation. Individual retinal photoreceptor cells have a dynamic range of about 100:1 but because photoreceptor cells adapt the offsets of their
operating ranges the overall dynamic range of the retinal system is
about 10^12:1 for global offset adaptation and about 10^5:1 for local offset
adaptation (D.C.Hood and M.A.Finkelstein, Handbook of Perception and Human
Performance, I:Visual Sensitivity, 1986, pp:5.1-5.66).
2. Ancient human environments were tough -- famines, diseases, droughts, warfare. Knowing how to solve problems using accumulated knowledge is a valuable survival skill. The older you are the more you tend to know. Genes for long lifespan could therefore have advantages both to an individual and to the community in which they live.
"the human retina has logarithmic response, thus you want a monitor where the mapping... is logarithmic as well. If you don't, you are not allocating the numbers in the most efficient way possible."
What you are saying totally contradicts numerous published papers and textbooks on visual psychophysics and colorimetry! The monitor used to display the stimuli in any reputable colorimetry or psychophysics experiment must either already have a linear response like a BarcoView CID421 or be calibrated (approximately) to have, a linear response.
"The steps are not Mach banding."
Please re-read what I actually said. I said the visibility of steps, not the steps per se, is mainly due to the Mach effect. It seems it's going to be difficult to convince you without taking you through a psychophysics demonstration with a properly linearly calibrated monitor. I recommend you read the excellent account given in Werner and Spillman, Visual Perception, Chapter 7: Perception of Brightness and Darkness.
"steps... equally hard to see. This requires a logarithimic mapping..."
I don't agree. Review any of the chapters in "Handbook of Perception: V Seeing", E.C.Carterette and M.P.Freidman (eds), Academic Press, 1975. Brightness perception is much more complicated than you are making out. To understand perception of complex stimuli, you really need to consider a range of phenomena including simultaneous contrast and border contrast effects not limited to Mach bands, as well as high-level perceptual processes such as form perception because they can alter apparently low-level processes like brightness perception based on neural feedback pathways.
"... the mapping from the real-world light levels to the numbers stored in the image file is not linear either, instead it is the inverse of the monitor."
No, most image sensors based on CCDs actually have a very linear response to light intensity. For a good introduction please read Barco's Color Theory Page, and note their comment about the linearity of scanners.
"... [logarithmic] human response and [exponential] monitor response... match far better than a straight line!"
I wish you were only joking. The dynamic range of the eye is orders of magnitude greater than any monitor, and the gamma is not even a constant in the eye which is luminance-adaptive both temporally and spatially.
And what does "match far better than a straight line" mean?
The VC firm is actually Linux Global Partners -- not to be confused with any VC firms containing "Linx".
Better workaround for cache-phobic websites
on
Deja.com Vu!
·
· Score: 1
Here's another handy workaround with Junkbuster which greatly improves the excessive latency of websites which set HTTP headers as Expires: 0(zero), Cache-control: as well as or instead of Pragma: no-cache. Here's a part of my extended patch (rest was cut for brevity) to fix the problem:
--- parsers.c.new Sun Nov 26 15:11:15 1998 +++ parsers.c Fri Feb 20 17:40:32 1998 struct parsers server_patterns[] = {
and to complete the patch just add the functions server_set_pragma, server_set_expires which can be identical to server_set_cache. You can also add extra functions for cleaning up any unwanted metas in the body.
Here's a workaround for Deja.com's no-cache
on
Deja.com Vu!
·
· Score: 3
A workaround for the excessive no-cache latency of Deja.com is to use Junkbuster with my patch to the file parsers.c as follows:
Slashdot Mirror
Well it's only possible... but with increased atmospheric CO2 and increased rainfall over the North Atlantic, you might be getting a lot less food than you do now (more info.)
One thing very few independent scientists dispute is the Rahmstorf bifurcation model of Atlantic thermohaline circulation has only two possible stable solutions:
The on-off switch is the amount of rainfall minus evaporation over the North Atlantic.
You might not like the possibility of this severely affecting Minnesota.
It's a pity attitudes to climate change research per se are so polarised. There was a paper in a climate modelling conference where the US author was only half joking when he used an amusing title on his paper.
That might be a very dangerous proposal indeed, not least because people might think it would work and actually test it without having any means to reverse it should it be harmful, but also because as per my previous post global warming might cause severe global cooling. (archive-mode proof link here)
What can global warming do? According to the widely accepted Rahmstorf bifurcation model of Atlantic thermohaline circulation, small increases above a threshold level in freshwater precipitation over the North Atlantic region can trigger a long-term Gulf Stream shutdown, a.k.a. an Ice Age.
A super-threshold increase in freshwater supply to the North Atlantic area is one of the most stable effects predicted in simulations by coupled ocean-atmosphere models of climate for any future increased levels of atmospheric CO2.
Would an ice age be so bad? Well, which of the two states below do you prefer?
A. Gulf Stream OFF image of US vegetation (Ice Age conditions)
or
B. Gulf Stream ON image of US vegetation (present day conditions)
What does an ice age do? On land, agricultural productivity in the major food producers like the US and N.Europe would collapse due to freezing temperatures. Reduced ocean circulation would also cut precipitation in other regions of the world, creating droughts. In the sea, the much reduced ocean circulation would be unable to replenish marine nutrients in vital high-latitude oceans, i.e., marine productivity and fishing harvests would also collapse.
I left out the word "only" as in: "land owner is only entitled in return to a reasonable fee", i.e., you may demand only a monetary fee.
You don't say which country you're asking about but in the UK and several other European countries too, a land owner is legally obligated through national laws to provide access to any utility company that requests land use and the land owner is entitled in return to a "reasonable" fee, which is usually negotiable based on a notional market rate. The UK legal term for such access rights is a "wayleave".
You might also be interested in learning written Chinese using a really excellent Chinese character tutor called "Hanzim" which is free GPLd software -- ~I posted more information about Hanzim here.
Hanzim, the Chinese character tutor, is also very good for learning the Japanese Kanji characters and multi-character combinations, which are based on traditional Chinese characters. Hanzim doesn't help with learning the Katakana or Hiragana, which are alphabetic not ideographic, but the alphabetic characters are simple to learn. There is also no Japanese pronunciation provided although it'd be easy to add to Hanzim's database.
Adrian Robert of UCSD has created Hanzim, a truly excellent tutor for people learning to read written Chinese. Hanzim is GPLd and has a nice GUI in Tcl/Tk 8.1.
I think the Hanzim project is an outstanding achievement. I wouldn't be surprised if the author starts getting a lot of press recognition for his work, and also one of the Free Software Awards for educational software. Afterall, he has created, with some help, a dictionary with over 6000 individual Chinese characters and over 18000 Chinese character combinations, each with English translations as well as a cross-referenced Chinese radical lookup facility.
The software is downloadable from Robert's Hanzim directory (http://zakros.ucsd.edu/~arobert/hanzim.html).
As a footnote, there is, of course, plenty of good commercial Chinese language tutorial software but Hanzim, uniquely, is under the GPL.
-- William
The Japanese word "docomo" is best translated as "everywhere", not as "anywhere".
In the UK you can opt out of paper junk mail by registering your name and address with the Mailing Preference Service. After I registered I got no more paper junk mail addressed to me. Occasionally I get junk mail sent to my address which have no name on them.
Freepost 22
London
W1E 7EZ
I was wondering whether there is a similar service in other countries?
The UUNET spammers collective is still being allowed to operate -- one of them tried a stealth port25 probe today but hit our firewall:
00:22:03 (EST) 04 January 2001: Port 25/smtp ACK/no_SYN connection DENIED from: 1Cust180.tnt38.det3.da.uu.net (63.44.201.180)
Coincidentally no doubt, this was quickly followed by the Harvard dialup scanners collective checking our netbios availability:
23:28:10 (EST) 04 January 2001: Port 137/netbios SYN connection DENIED from: sfp220-198.harvard.edu (128.103.220.198)
Someone please tell me UUNET and Harvard are doing something to stop these guys.
The Anchorage quake (1964) was Magnitude 8.5
Have they included the massive extra cost of building an earthquake-proof tunnels? The whole region between Alaska and the Siberian Peninsula is located over major fault planes. Geologist's have concluded there was a massive quake a few centuries ago that was much bigger than 8.5 and possibly as big as 9.7 (more than one order of magnitude more energy than an 8.5)
Nobody really knows how big the quakes can get out there.
JPEG2000 copy control is described in the document JPEG2000 requirements and profiles version 6.3 under section 5.8 Security, p.12, published in July 2000.
The JPEG2000 copy control requirements were apparently added only recently as part of the application requirements but the wider goal is to "ensure that appropriate tools and technologies are adopted in JPEG2000 standard to fullfil [the requirements]". They are not mentioned at all in the full draft ISO/IEC specification Final CD published on 16th March 2000.
1. It's too early to criticise The Gimp for not supporting JPEG2000 because the JPEG2000 image standard has not been finalised. To quote the Final CD from the official JPEG2000 website,
- "anyone implementing software according to the description available in this FCD, risks not being compliant with the final JPEG2000 International Standard (IS), which is due to be published some time in 2001 as IS15444-1.
So don't expect any projects either to implement JPEG2000 yet or to be able to get JPEG2000 patent licenses (see below) to do so yet.2. Another important issue (stated here) is that the royalty-free fee-free JPEG2000 patent licenses may apply only to conforming implementations. One such requirement of conforming implementations is to have copy control implemented including methods
It is not clear how the patent holders will interpret these requirements for open-source implementations that want to use a GPL license (as for Gimp) which requires the whole software to be modifiablefor any purpose, potentially in ways which could violate the conformance requirements.
--William
I started this thread talking about anti-aliasing of color text images in the context of using uncalibrated CRT monitors. Keith's point, and mine too, was that for anti-aliasing of color text images to be correct, color interpolation must be computed in a linear intensity space. At this point, the topic was still anti-aliasing of color text images. It was therefore quite a surprise when you replied to Keith Packard's followup comment saying,
- "[it would be wrong]
- to change the DAC converters [aren't DACs always converters?] because adjusting the monitor to be
'linear' would spread the samples at the dark end out so much that the steps
between them would be easily visible, while compressing the samples at the
white end so close together that thye cannot be distinguished"
You totally misunderstood Keith's point, and went along a tangent talking about brightness steps which is not obviously relevant to anti-aliasing of color text images. Let's stay on my topic of anti-aliasing in a linearised intensity space.As you likely know there are two obvious ways of getting an approximately linear intensity space for a typical CRT connected to a graphics card with programmable DACs:
- 1. Adjust the DACs to get a linearised intensity response
by programming them with inverse gamma lookup tables.
Method 1 is fast but has the disadvantage, acknowledged in Keith's comment, of not producing a linearised intensity space for images which already have gamma-corrected gray-levels. Method 2 is slow but can be applied only where necessary, i.e. only to gamma-uncorrected images.2. Adjust the gray levels in each color channel of an image by mapping gray-level through inverse gamma lookup tables.
Regardless of which method is used, the result is a linearised intensity space. Doing anti-aliasing in any other intensity space such as the exponential intensity space typical of an uncalibrated CRT would be wrong.
Oops, I meant the Chevreul illusion, not Mach bands! The Chevreul illusion is the apparent increase in brightness contrast between adjacent steps in a "staircase" image containing steps of uniform gray-level which increases from each step to the next step. The reason why any steps are visible in gray-level staircases with more than about 30 steps is the Chevreul effect. If large black gaps were inserted between adjacent steps in a staircase image you would not be able to see any brightness difference between adjacent steps. That's because the gaps between the patches prevent border contrast effects such as the Chevreul effect which are necessary for us to estimate relative intensity accurately -- do an experiment and see for yourself.
Linearity of digital cameras:
- "the circuitly in all digital cameras forces that linear output
through a lookup table to convert it to a nonlinear value"
That's not true. I've done spectrophotometric calibrations of many devices including famous-brand CCD-based camera and scanner systems; none of them has had any significant non-linearity in the mapping of intensity to gray-level. There are rare, special-purpose CMOS-based sensors with deliberately non-linear analog responses like Mead's artificial retina but they are not used in any consumer camera systems. The old "plumbicon" cameras based on vacuum tubes were the last common imaging devices to suffer a severe non-linearity -- an exponential response similar to CRTs due to their use of phosphors.You're right you need to apply a logarithmic mapping to gray level g'=log(g) for a CRT display but your explanation is wrong. The real reason is because a display should have a linear output light intensity with respect to both input voltage and gray-level.
A logarithmic mapping linearises the intensity space of a CRT but it does not equalise the contrast differences in a gray-level staircase, which actually requires an exponential intensity space. To see why you'd need an exponential space for contrast equalisation, consider the gray-level staircase image with y as absolute intensity, dy/y as relative intensity between adjacent steps (aka brightness contrast) and x as the gray-level. To get the relative intensity to be a constant we have dy/dx=y.C implies x=log_e(y)/C = log_K(y) where C=log_e(K), so y=K^x. In plain English this means a staircase image is contrast-equalised only if absolute intensity is an exponential function of gray-level. Of course a CRT without gamma correction already has absolute intensity as an exponential function of gray-level, so gray-levels don't need to be changed by any gray-level mapping. A logarithmic gray-level mapping would make brightness contrasts decrease across the staircase.
Your dynamic range is too small. The eye actually has a massive 10^12:1 global dynamic range. The pupil doesn't account for much dynamic range; the range of pupil diameter is 2-7mm giving only a 38:1 range of light attenuation. Individual retinal photoreceptor cells have a dynamic range of about 100:1 but because photoreceptor cells adapt the offsets of their operating ranges the overall dynamic range of the retinal system is about 10^12:1 for global offset adaptation and about 10^5:1 for local offset adaptation (D.C.Hood and M.A.Finkelstein, Handbook of Perception and Human Performance, I:Visual Sensitivity, 1986, pp:5.1-5.66).
2. Ancient human environments were tough -- famines, diseases, droughts, warfare. Knowing how to solve problems using accumulated knowledge is a valuable survival skill. The older you are the more you tend to know. Genes for long lifespan could therefore have advantages both to an individual and to the community in which they live.
Which do you prefer?
A. Gulf Stream OFF image of US vegetation (Ice Age conditions)
or
B. Gulf Stream ON image of US vegetation (present day conditions)
eating too much is as base an instinct as suckling
Since when did suckling become a contemptible activity?
What you are saying totally contradicts numerous published papers and textbooks on visual psychophysics and colorimetry! The monitor used to display the stimuli in any reputable colorimetry or psychophysics experiment must either already have a linear response like a BarcoView CID421 or be calibrated (approximately) to have, a linear response.
Please re-read what I actually said. I said the visibility of steps, not the steps per se, is mainly due to the Mach effect. It seems it's going to be difficult to convince you without taking you through a psychophysics demonstration with a properly linearly calibrated monitor. I recommend you read the excellent account given in Werner and Spillman, Visual Perception, Chapter 7: Perception of Brightness and Darkness.
I don't agree. Review any of the chapters in "Handbook of Perception: V Seeing", E.C.Carterette and M.P.Freidman (eds), Academic Press, 1975. Brightness perception is much more complicated than you are making out. To understand perception of complex stimuli, you really need to consider a range of phenomena including simultaneous contrast and border contrast effects not limited to Mach bands, as well as high-level perceptual processes such as form perception because they can alter apparently low-level processes like brightness perception based on neural feedback pathways.
No, most image sensors based on CCDs actually have a very linear response to light intensity. For a good introduction please read Barco's Color Theory Page, and note their comment about the linearity of scanners.
I wish you were only joking. The dynamic range of the eye is orders of magnitude greater than any monitor, and the gamma is not even a constant in the eye which is luminance-adaptive both temporally and spatially.
And what does "match far better than a straight line" mean?
The VC firm is actually Linux Global Partners -- not to be confused with any VC firms containing "Linx".
Here's another handy workaround with Junkbuster which greatly improves the excessive latency of websites which set HTTP headers as Expires: 0(zero), Cache-control: as well as or instead of Pragma: no-cache. Here's a part of my extended patch (rest was cut for brevity) to fix the problem:
- --- parsers.c.new Sun Nov 26 15:11:15 1998
- { "set-cookie:", 11, server_set_cookie },
and to complete the patch just add the functions server_set_pragma, server_set_expires which can be identical to server_set_cache. You can also add extra functions for cleaning up any unwanted metas in the body.+++ parsers.c Fri Feb 20 17:40:32 1998
struct parsers server_patterns[] = {
- { "cache-control:", 14, server_set_cache },
- { "pragma:", 7, server_set_pragma },
- { "connection:", 11, server_set_connection },
- { "expires:", 8, server_set_expires },
-char *server_set_cache(struct parsers *v, char *s, struct client_state *csp)
-{
- return(crumble(v,s,csp));
-}
A workaround for the excessive no-cache latency of Deja.com is to use Junkbuster with my patch to the file parsers.c as follows:
+++ parsers.c Fri Feb 20 17:40:32 1998
@@ -27,26 +27,20 @@
{ "from:", 9, client_from },
{ "cookie:", 7, client_send_cookie },
{ "x-forwarded-for:", 16, client_x_forwarded },
{ "pragma:", 7, crumble },
Or if you don't follow the diff, just add the pragma line above by hand. Just recompile and install it.